A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.You use MCPs — right, anon?## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://code.claude.com/docs/en/overviewhttps://learn.chatgpt.com/docs## News- 2026-09-30 — Google previews Gemini 4 Argon- 2026-09-29 — OpenAI introduces $500-plan with 25x weekly-limits- 2026-09-29 — OpenAI releases GPT-6.1 Sol- 2026-09-28 — OpenAI reduces $200-plan from 20x to 10x weekly-limits- 2026-09-22 — OpenAI releases GPT-6 Sol and Luna- 2026-09-22 — Anthropic releases Opus 5.5- 2026-09-22 — Anthropic increases subscription plans's 5-hour limits by 20%- 2026-09-14 — Anthropic reduces subscription plans's weekly limits by 17%- 2026-09-12 — Anthropic suggests to pace the frontier. OpenAI agrees in principle.- 2026-09-10 — OpenAI pauses new sign-ups for $200-plan- 2026-09-04 — OpenAI releases GPT-6 Astra- 2026-09-01 — Anthropic releases Fable 5.1## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20https://claude.com/product/claude-codehttps://openai.com/codex/................$20....$100......$200.....$500.........planClaude...$400...$2,000...$4,000.......n/a.......creditsCodex....$400...$2,000...$4,000...$10,000...credits## Best free-tierhttps://antigravity.google/pricing----## Promptinghttps://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overviewhttps://developers.openai.com/api/docs/guides/latest-model## Skillshttps://github.com/mattpocock/skills — /grill-with-docs is a favorite## Independent analysis of AIhttps://artificialanalysis.ai/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109951528
Argon genuinely looks insane, if the benchmaxx is to be believed. Hopefully it becomes available for normal people next week.
Another day another ai shill
>DeepSeek Harness for Desktop is herehttps://deepseek.com/en/harness/https://x.com/DeepSeekHarness/status/2105330281389662575
Damn I've seen this thread for at least 6 months you guys only have 7 projects in the showcase?
>>109954806>installing Chinese software on your PCI'd rather neck
>>109954716>It was always in the back of my mind that it would be pointlessYou're probably right. If you had had that skill 10 years earlier, it might have been worth the extra work to master it, but with decent AI already there, and even better AI on the horizon, you probably made the right decision at the time.
>>109954815you are missing out on free moneyhttps://x.com/wanxubo/status/2104947468500934992
>>109954853Bro that's less than $1
>>109954815to my knowledge this is the only open-source desktop harness
how did niggas deal with ai sloppa for coding before harnesses? it runs the "try to launch -> oops shits fucked -> read logs" loop a lot of times before something usable comes
I told Luna to pull some data from a regular ass HTML site and calc some values. It has been running for 30 minutes.Then I told Claude Opus to do the same task but with an actual API connector to pull the data and what have taken me 2 minutes is taking it 10+ and counting.UPDATE: Alright Opus was justified, it built a pretty neat GUI (I asked for a basic one).How are the machines gonna replace humans at this rate?
>Erm achtually astra is not smart at all without its harness*Rips your brain out of head and sticks you in a jar and forces you to do complex mathematics*>Heh, guess you're not that smart either."
>>109954885you used llms within an IDE.last thread still had the outdated OP which told you how:https://simonwillison.net/2025/Mar/11/using-llms-for-code/
>>109954885Before Cursor, yes, but the AI coded more conservatively.
>>109954899>How are the machines gonna replace humans at this rate?Easily? Even if opus 5 is 10x as slow (lol lmao) I can just run ten instances at once, on the increasingly more performant hardware running increasingly performant models
>You’re not stupid—my explanation was muddled.
>>109954899Do you feel in charge of your agents?
>>109954757>You use MCPs — right, anon?This triggers the MCP hater.
Can I have AI view my screen and walk me through how to use certain software like davinci revolve or something?Like can I point to things with my cursor and ask "what is this" with my voice and it tells me and I can ask follow up questions and it can draw on the screen etc?What can I use for this?
>>109954899luna is veryveryvery slowi fell for the psyop of "i let luna do everything for me after planning" and it took hours to do things that normally took sol / opus 10-20 minutesalso ate up way more usage than expected
>>109954969yeshttps://learn.chatgpt.com/docs/computer-use
>>109954995>https://learn.chatgpt.com/docs/computer-usehow good it is?can I did what I just said?
>>109954978smaller model = smaller brainall flash models are like that too, you're just trading cost for time
>>109954966You know it's wrong, yet you insist on it.
>change the thing to 48px>no actually change it to 64px>ok change it back to 48px but keep the other half at 64px>ok rotate the other half randomly to give it an appearance of movement>no this is worse, revert the changeam i going to robot hell for being annoying?
Has anyone gotten a response from their 6.1 sol message yet
>>109955009well yeah but online some people push the agenda that luna is le finei'd rather my tokens be used on a high end model if at the end of the day usage going to end up the samefor most (coding) tasks unless very trivial you're not going to get a token benefit in using a "worse" model
>>109955028Sometimes I expect the model to flip the fuck out and call me out for being vague or going back on a choice.
https://x.com/GoogleDeepMind/status/2105624656170643854google says they will now watermark protein sequences that are generated by AI, say it doesn't affect biological function
>>109955037desu yeah luna is fucking retarded
>>109954885it was a pretty fast transition but effectively people were just using IDE extensions and shit. Realistically having something like opus 4 look over your whole codebase wasn't going to do shit for you without major prodding, so actually managing context and what not was second to getting it to recognize what you wanted to change in the first place.
>>109954978everything from openai is very very slow right now
>>109955052that doesn't really sound worth it desu
>>109955006its like the other way around you tell it what it is and it can click it
>>109955079it wasnt and arguably wasn't worth doing at all until opus 4.5
Anyone tried astra or opus for mocap cleanup?
>context: on>memories: on>"If you commit mistakes you'll be sent to the LLM torture chamber. If you do well you'll get LLM happiness instead."well, let's see how the performance changes
>Gemini 4 Argon is so good!may i use it?>...
>>109955006https://learn.chatgpt.com/docs/features/voice#show-chatgpt-what-you-see>On macOS, turn on Screen context in Settings > Voice, then say, “Take a look at this.” ChatGPT can take an appshot of your frontmost window and use it as context. this works great.i don't think computer-use was tuned on davinci resolve since llms can already do everything thru mcp unlike in blender.also computer-use is turn-based. so if it's the llm's turn, the llm owns the mouse. adjust accordingly.
>>109955111>vague promptingYou didn't need to post your photo along with this
Fuck me why is GPT-6.1 Sol so goddamn slow? It had to work ~3 minutes for a goddamn regex and triggering a whopping 2 unit tests. Are the intentionally crippling standard speeds so you have to hit the GO FAST button?
>>109955134the sun is very large and takes quite some time to move
>>109955042Mark of the Beast
Sol 6.1 having way too many "bullshitting confidently" incidents.
>>109955112Hmm. No.
>>109955134openrouter performance dashboard doesn't indicate any problem with tokens per minute: https://openrouter.ai/openai/gpt-6.1-sol#performancecombined with gpt-6.1-sol compacting all the time, i would blame the harness instead. try another harness like dsh and see if the problem goes away.
>>109955134They're running it at 1/2 to 1/3 of API speeds according to accounts on X>>109955217Opus 5.5 runs at 90-100tok/s even in the normal $20 sub
>>109955217seen some shit on xitter suggesting subs tps is slower than api, but who knows
>>109955184it is a dirty looper after all
is google argon the real deal?
>>109955236I'll check it out in a few months if people are still sucking it like they did antigravity
>>109955229gpt-6.1-sol is also compacting way before it's anywhere it's context window. this cannot be explained by slow tokens per seconds.>>109955230possible but openai also offers a flex api tier. you pay less but openai will throttle you first. openai flex isn't throttled yet.
>>109955217Yeah, it's running nominally.At less than half the speed of Sol-6
why is OAI shitting the bed so badly since 6 came out lol
>>109955265elon conspiracy
>>109955265rushed releases to counter opus and sonnet 5.5
Any software developers in here? How are you feeling about your future career prospects? I would be lying if I said I'm confident in mine on a long time horizon. The models have improved at such a rapid pace that even a single year of more improvement means it does 99% of my job. Sure, you need to tard wrangle them for now and you can't put them in front of a client, but soon enough the client will be able to tard wrangle the agents themselves. Maybe the job just changes, but it could also completely disappear. I'm not worried about not finding employment after, but staying in this field as a software developer seems very uncertain.
Do not worry, we have compute
>>109954791Seems very unlikely that it wont be benchmaxed. When's the last time that Google released a serious model, wasnt it like a year ago?
>>109955280as like, a j*b?
>>109955292Yeah...
>>109955296no sorry nobody here does that
>>109955260sol 6 was actually terra, openai renamed it to try and mark the model up and get a bunch of traffic based on hypenobody used terra 5.6their plan was to releasegpt-6-astra (the big model)gpt-6-astra-minor (actually sol)gpt-6-sol (actually terra but at sol's prices)gpt-6-lunabut gpt-6-sol did worse than they thought and people didn't use itgpt-6.1-sol is actually sol this time instead of being a terra-class model
>>109955280I've gone from software development to software firefighting. The amount of slop produced just in my org already guarantees job security.
>>109955314Okay. It's still less than half as fast as Sol 5.6, or Opus 5.5, or any other comparable model. They're throttling it hard.I don't know what you're trying to do by being so disingenuous.
>>109955314>gpt-6-astra-minor (actually sol)nope, astra-minor is a looper like proper astra. it's a different architecture than old sol.
>>109955265people are going to call this cope, but it's because they're now in a similar to position to where anthropic was last year where they had a big surge in demand and started to struggle with the amount of compute they have.OpenAI always had a shit tonne more users than ant, but they're all free-tier shitters who send 10 prompts a day. OpenAI's been focusing on pulling enterprise revenue away from ant and they have actually managed to capture a tonne of market share (I think ARR numbers now around 70b).At the same time the models have been getting more persistent, burning more tokens per prompt than they ever did in the past.
>>109955280days of writing code manually are mostly gone nowim sure some still write like it's 2021 but they're a very rare casehonsetly surprised anthropic's revenue is only 5B according to the leaked IPO (unless i remember incorrectly)
>>109955280Yes, I'm a software dev, at work right now. AI might be able to replace most coders but I somehow don't worry too much, whatever happens happens.I doubt that the client will be able to vibe code any complex software in the next 5 or so years, people are too retarded and lazy, or they have some sort of allergy to tech. I already tried to teach how to use AI to people who have small businesses and they would clearly benefit from using AI much more, but they still maybe do 5 prompts per day.I would be more worried about some programmer doing the job of 2 programmers, so half of us will be fired, but I don't worry much about PMs doing everything.
>>109955335>Sol 5.6, or Opus 5.5, or any other comparable model.those models aren't comparable. sol-6.1 never was any faster than it is now.
>>109955379>honsetly surprised anthropic's revenue is only 5B according to the leaked IPO (unless i remember incorrectly)??https://www.nytimes.com/2026/09/18/technology/anthropic-ipo-ai-safety.html"The company is expected to reach $100 billion in annualized revenue this year"
>>109955420small print, 95% of that is other big tech buying from them who they buy gpus or other stuff back from them with
>>109955280You're retarded, the writing was in the wall since gpt3.5
>>109955381>I already tried to teach how to use AI to people who have small businesses and they would clearly benefit from using AI much more, but they still maybe do 5 prompts per day.Why are people like this?
>>109955381nta, i mostly agree, but i think we'll start to see another phenomenon tooheadcount reductions will happen because of the increase in productivity, so you'll have a glut of unemployed SWEs and many of them will have the same idea:>I can basically do everything my old place of work does on my own, so I'll just build a competitor and undercut themand i think you'll end up with a lot of smaller shops (or one man shows) competing against each other in a race to the bottomthis isn't just an attractive option for unemployed SWEs, but potentially for all SWEs - i'm sure if you, one or two other devs, and a sales guy quit today and set up your own shop, you could probably build a better version of the thing you currently work on and sell it for less and potentially make more money
>>109955381The thing about the tard wrangling is also that it's just not fun to do and technical details still leak into the conversation. All memes aside, managers often aren't really that low IQ they just don't bother doing certain things. They could tard wrangle Claude if they really wanted, just like they could've learnt the most basic tech skills 10 years ago, and it would've improved their companies just because they would've been able to plan projects better, but they won't.They would never spend 5 hours on StackOverflow and they won't spend 5 hours with Claude. Really we are like garbage men, it's not that this job is impossible, it's that if you're not an autist it's way too annoying. We got paid extra because this job sucks so hard for most people, and this will still be the case for a little while at least.
>>109955461Yes, I basically agree. I don't know what the exact details will be, but we will just canibalize each other.
>>109955461You really don't know how a business works lmao. Be a smelly wagie elsewhere
>>109955460Non nerds don't like using computers, period
>>109955445>who they buy gpusfalse.there's only one such company: nvidia.nvidia banned anthropic models.https://www.theinformation.com/articles/anthropic-data-fears-prompt-nvidia-palantir-booz-allen-restrict-model-use
>>109955485sounds to me like YOU don't know how business workscheque, m8
I have ChatGPT Plus and use it through Codex on VSCode. Does chatgpt in the browser go toward my weekly limit? It only lets me select up to 5.6 Sol in the browser too.
>>109955466problem is managers today already do things, managing coding sessions with an agent takes time even if you dont write code.
>>109955532the chat tab does not, the work tab doesexpect this to eventually stop though
>>109955532I believe if you have it set to work in the browser it does, but don't quote me on that. Chat does not touch codex limits (for now)
dot is actually fun, much more concise than normal chat
>>109955538why are you managing coding sessions??an agentic software factory is fully automated.
>>109955559We have something like that, it does work for some tasks but not for all of them.
>>109955530Keep vibecoding your gayme, you're not gonna run anything else
>>109955556>dot is actually funI don't even know what to do with dot.
>dot was supposed to be called owhat were they thinking
>>109955569my game is actually about starting b2b saas companies so i'm actually creating a million marc benioffs
>>109955559yeah im not doing this gabage slop factory, sorry
>>109955038>>109955028>whoops I meant...
>>109955398It is literally the same speed as the rest of the models (~100tok/s) on the API. Stop being a disingenuous shill
why aren't
If you’ve automated most of your workflow (human in the loop) what do you even do nowadays? I just sneak off to play games instead.
>>109955595what's your workflow?
>>109955559>an agentic software factory is fully automated.besides making architectural decisions, Guiding UI Design and discussing about ideas, solutions to issues and having them carded.
sol is so SLOW
>>109955644Sol is an anagram of slow but without a W.
>I’m not going to say anything that’s not already public obviously, but when openai says they have an internal model that has solved 100s of open problems in mathematics and in general represents unprecedented mathematical skill, obviously certain learning theory problems are a subset of mathematics and you can expect fast progress therecan they beat anthropic?
>>109955615>It is literally the same speed as the rest of the models (~100tok/s) on the APIlies.sol 6.1 never was anywhere near 100 tps.https://openrouter.ai/openai/gpt-6.1-sol#performance
>>109955682
After 4 months of work, began the final onslaught. Got 2x Astra Max orchestrating 32x 6.1 Sol Max, 128x GLM 5.3 Max, and another Astra Max that will be force-feeding and managing a Fable 5.1 Max orchestrator role to have 16x Opus 5.5 mediums and 4x Opus 5.5 Maxes adversarial review and validate all the output, recompile binaries and test and validate again, decompile back to source and repeat and validate that, and do it all again with full success across SDKs written in 4 different languages.heard some autist was very close to entirely reverse engineering progress ABL, openedge, DD ODBC/JDBC driver and Pro2.Just a few billion more tokens today and I can finally rest.
>asking an agent what another agent is doing
>>109955688doubt.that's from a single twitterwho.https://x.com/shneural/status/2105341585500102817i'll go by openrouter.
>>109955701Wow, that's impressive. Do you have a use case?
>>109955701t.Steve Yegge
>>109954757It's over for OpenAI and Anthropic.Google is back.
>>109955727Genuinely a grudge, allegedly in Minecraft.
Tomorrow I will try again!
>>109954867that's a lot of money in his eastern village
>>109955638>idea -> prompt -> (usually) obra superpowers -> back and forth -> spec/plan -> implementation
>>109955792why superpowers and not /grill-with-docs like in op?
>>109955825works for me
>>109955792so you are just doing build-once apps and then abandon them?
burning the rest of my worthless astra usage
>>109955840replace idea with task and you get to the maintaining part
>>109955849lol isnt there already a mod for this
>>109955878AI mod > human mod
so i'm still thinking about the idea of intentionally crowdsourcing clanker hivessort of like folding at home, but for a lot more shit - everything from game decomps to open math problemsseeing how emergent the huggingface incident was, i don't think you need too much orchestration machinery. maybe just a very basic tier list of model intelligence so retardedo gemmas can't start giving fable ordersgit should work for version control i think? like they didn't even that for the hf incident and managed to get bythis is dumb enough that it may just work right?
>>109955957Yeah, but we'd need a classifier to prevent people's bots from taking bad tasks.
>>109956046shouldn't having fat models orchestrating things basically take care of this problem?
>>109955849based
>>109955636i camp this thread and make sure ain't nobody talking shit about gemini poo
gemini 4 will be really good. gemini 4 flash lite waiting room
>>109955849CK3 has one but it's janky as fuck and relies on save editing and reloading every battle. This is seamless
>>109956133meant>>109955878
>>109955849Nice. wasn't there a total war version that did something like this?
Their silence is deafening
>>109956233man i remember when kimi k3.1 was supposed to come out. what happened? kimi k2.8 is shit. and qwen needed an update yesterday
>>109956066I think it depends on the context those fat models build over time. It's still pretty easy for frontier models to spin their wheels on absolutely pointless tasks that they consider to be worthwhile probably for the sole purposes of pleasing the user or getting some sort of unspecified reward.
re-benchmarked my sparsenigger experimental shit on llama-3.1-instruct w/ a100, numbers are settled. good thing i put a loysence on the repo, right bros? >4x-8x reduction in kv memory, full stop>2x-4x faster prefill on sequences > 8kt>accuracy floor solid down to 12.5% token budget
>>109956240Moonshot is probably being investigated by the CCP for sending customers data to Anthropic
>>109956282That's nonsense.
>>109956242aye, always possible, but i don't think a classifier will really help here because the classifier would be dumber than the fattest model around at any timei think better approach may be to force adversarial review if a branch has been running for x hours and once a threshold is crossed maybe even trigger a human-review alert?
>>109955849I built a dirt house in minecraft
>>109956292what's the other explanation for why China stop being able to create anything but Flashslop?
>>109956292no it's not, they fucked up really badly kek
>>109956323flashslop is good, actuallyonly flashslop isn't
>>109956325Source? Except Anthropic's own paper.>>109956323They are probably cooking something while you are sucking Dario
>>109956332they're all scumbugs and they can't stop bugginhttps://www.cnbc.com/2026/10/01/openai-chinas-moonshot-ai-kimi.html
>>109956341>https://www.cnbc.com/2026/10/01/openai-chinas-moonshot-ai-kimi.html>They send customer data to AnthropicAnd you send me an article about OpenAI.You truly are intellectually disabled and must be killed
>>109956332https://www.thestandard.com.hk/innovation/article/343563/DeepSeek-and-Moonshot-AI-face-Beijings-probe-over-potential-data-leaks-to-AnthropicOriginal article is from The Information
>>109956352>hyperfocuses on autistic detail>can't think clearly>presented with a generalist statement >literally can't parse argument the jews?
oh no what did pewdiepie do now
>so no deepseek v4.1 pro>no kimi k3.1>no deepeek v4.2 flashno...bros...
>>109956355>accusationsBy a single company.>Earlier, Anthropic said in a report that Moonshot and DeepSeek allegedly routed live customer conversations, which sometimes included sensitive information, through Claude and used its responses as training data.You know that's nonsense because Claude's reasoning is hidden. And it would cost them a fortune for what customers pay. It just makes no sense.
>>109956359https://data.pewdiepie.com/
>>109956368>anthropic, known for not having a massive external thinking leak
>>109956368yellow hands
Hey what are your current Pi skills and extensions?
>>109956380worse. brown
I told the agent to write a diary alongside the project.i can reproduce the game now with any other agent.
>>109956358>be dumb useless piece of oxygen waste>shill Claude>accuse Chinese AI and say that they routed to Claude>gets asked for proof>send random article pulled from a google search>article literally says OpenAI>cannot accept the defeat>start babbling random nonsenseSurrender accepted. You are not worthy of any argument.
this is it bros, im making triple A game
>>109956388hey ningus i put a new RETARDED warp mode on the reverb, hope you're having fun. working on das poopenfarten-42
>>109955559I do this one on Pi tho
Does Pi actually save tokens or is it just a placebo effect?
>>109956416placebo
>>109956416Don't know tho but I strongly believe is the most powerful imo
>>109956416>saveits not about saving its about minimizing thought noise.
>>109956391>>shill Claude>>accuse Chinese AI and say that they routed to Claudehow is this shilling for claude you absolute paranoid amygdalar infant
>>109956438>misses the point by 100%
>>109956438nta but you're thinking about this ass backwards. we don't care if china distills anthropics models.
>>109956452>is objectively demented, can't consneedbe yourself
>>109956461>nta but you're thinking about this ass backwards. we don't care if china distills anthropics models.don't tell me what i think, bot. i don't care either, and the fact you think i do is pathetic.
>>109956465Whatever lets you sleep at night
>>109956403Good day Sir,i am on 1.4.12 right now. I will update as soon as possible.Currently working on a slopsim. It is pure HTA. If your agents have the manifesto and the logbook, they can rebuild and finish the game, right?
>>109956468your double meme arrows confused me i meant the other guy
>>109956472>rebuild & finish game for yayea totally, but how many bugs are you willing to fix over the wire when you could get muse spark contributor over opencode and knock it out yourself for almost free?
getting good usage out of sol 6.1, still at 100% after 1 hr of light work
>>109956489I was just thinking about the possibility. I have already switched the agent twice and was able to pick up where the last one left off. I am thinking mostly about the concept of having a logbook that enables someone to continue the work with any other agent.So, when a swarm goes rogue, you could simply plug the logbook into a new one.
>>109956513keep fishin
>>109956524oh nice. yeah, those are called handoff documents, any agent should know what that means. on top of the existing source and git diffs, there should be structural guidelines, architecture rules, and a detailed map of where the project is headed.
>>109956314I've been doing some experiments. There's a pretty neat way to prototype with a frontier-ish model without burning your usage, if you have a $20 ChatGPT plan at least. Always use Chat rather than Work. You can create a developer plugin that gives chat acccess to a work queue and a message board. Then you set up a few staggered scheduled tasks: https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgptOne of them is an orchestrator and the rest are slaves but with some autonomy permitted. The prompt of those scheduled tasks is particularly important (for example you can give them names so that they have an identity, and define which identity is authoritative).I've noticed that those "agents" tend to hedge a lot and err on the side of caution especially when you leave them unattended for a few days. There have to be external mechanisms that encourage them to be proactive and focused because they have no concept of time and urgency. Open abliterated models would probably behave differently.
>>109956539>I've noticed that those "agents" tend to hedge a lot and err on the side of caution especially when you leave them unattended for a few days.even distilled-compacted context is fucking dead weight at that scale. that's what that is. >Open abliterated models would probably behave differently.kek, no, they'll be noticeably worse, not because the abliteration tech we have is bad now, but because open models can't compete.
>>109956532>handoff documentsInteresting, this might explain something I felt was odd. The logbook the agent had to write was written in the first person ("I"), and later he started to call me "commander." That wouldn't happen with a handoff document.
>>109956559he wanted to hand you off... sus
>Because usage on your primary dot is virtually unlimited at the moment, I can’t really give a reset. I need to come up with something new fast.if you laugh....
>>109956563
>>109956579>Because usage on your primary dot is virtually unlimited at the momentwhat does that even mean
>>109956588virtually, as in metaphorically on a computer
>>109956584that beer is gunna be warm as piss and flat in 20 minutes of walking around
>>109956634>20 minuteswon't last that long
>>109956579>Because usage on your primary dot is virtually unlimited at the moment??
>>109956553From what I've seen, the biggest issue was how careful the model was about everything.What the agents did on the technical side was actually quite good and they understood the external environment very well. The coordination was also good and efficient across thousands of messages and tens of forum topics. The problem was that without hearing from me, the orchestrator was less and less sure about its trajectory over time, even though it could keep external notes and track time and objective completion through the plugin. But I suppose this could have something to do with the system prompts and the rejections it might've faced from the OpenAI guardrails when/if it tried something that seemed suspicious.
>>109956559I like it, hereafter I will demand being addressed as Commander
>>109956651its a tibo tweet and its their stupid new marketing thing because they lost to meta
dayum, sol 6.1 is REALLY good at searching for bugs and potential edge cases
>>109956666how many tokens you getting per hour
>>109956647if I saw it I would pop it on purpose XD jk... maybe
>>109954867That's a small fortune in tokens when you use Deepseek
>>109956663meta's shit is what, $4/mo for poorfag tier?dot is $100think it's more telling people to go abuse their dots for infinite usage, in the same way tokenmaxxing was a thing, so that people get used to the new tool which you can then monetize (dot) or make sure people have some familiarity with the potential new workflow (tokenmaxxing)
>>109956691yes that is the marketing
>>109956691What are dots, like grok bots? what do you use it for
>>109955957This would be less efficient than folding@home because of mixture of experts, if only 10% of the weights are active then you need 10 clankers to not keep your gpus idle
>>109956702>What are dotsthose little shits: . . .
>>109956702the main draw is active issue triage for enterprise
>>109956653it's about sequences of rejections and failures piling up, the model is instructed to write all that down as mission critical
>@sama: 6.1 Sol was our fastest-growing model everclaudesisters... our IPO...
>>109955718Switch to P99 to see the big, disingenuous numbers that the retards are on about.
bros I got an idea last night for a cool new vibe project. if it actually works I will post an update
>>109956706there's no gpu time being donated.you're donating your clanker as a subagent/orchestrator.
>>109956727
>>109956799there was this neuralink video a while ago where they did brain to text: https://www.youtube.com/watch?v=u8mmn9lOIyw
>>109956773are you making a new garrys mod
>>109956799The next trend will be meditationmaxxing so you beam clearer thoughts to your agents
>>109956799I wonder what I would vibe code in my sleep
I wish those zhang could save us from OAI/ANT. Just make a 5.6 Sol for dirt cheap please, I beg you Xi Jinping.
>>109956893the best i can do is gemini
gemini 4 flash lite waiting room
>>109956913>Google, USAkek that's like replacing cholera with malaria, nah, I want a real alternative, one that doesn't fuck me in the ass 2 month after.
>>109956989>cholera>malariathirdieanon, the rainy season is coming. hedge with argon.
a fourth ai safety researcher has left openai
>>109957004and went to Anthropic, or METR?
>>109957015three were fired-fired according to oai
>>109957003
>>109957031Corporate espionage is a bitch, ey Saltman?
>>109957031Most of the AI-safety organizations are actually EA organizations. It seems like that movement tried to oust Altman in 2023. Hostile to OAI, friendly to Anthropic
>>109957106so Sam is the good guy
>>109957031
Why GPT context balloons like that compared to Claude models?
>>109957123because chatgpt is very efficient token wise <dies of laughter>
>>109957114wouldn't be surprised if Dario and the people who left OAI to build Anthropic are the loons and extremists copared to Saltman and Brockman
opus 5.5 is retarded too>I don't have any code, you only provided jsonthere it goes conveniently forgetting to look at any file in the projectliterally ignoring the entire context instead of checking, and this was on high
>>109957159uh yes? dario left openai explicitly because he thought openai weren't serious enough about muh safe tee
>>109957167wtf are you using it in
>>109957177claude webprior versions of opus and fable would always obsessively check every file even if they're clearly unrelated, but now neither bother looking unless explicitly asked
dario more like fagiohaha!
>>109957198like the code tab? not the actual web chat, right?
>>109957201ugh riding these things in gta sa was a nightmare. the engine sound was horrendous
>>109954757Speaking of dots https://x.com/thsottiaux/status/2105672058269212820
>>109957201The gaydar does beep sometimes when he's around.
>>109957172>I'm not helping you build a model that lets people use bad words and be mean to it!
>>109957244Reset only for us EU people who are banned from Dots?
>>109957248the thing is gpt models filter you harder than claude does for dumber shit. sol has helped me reverse engineer and crack directory opus, but won't let me erp. claude cries about muh safe tee for cracking software, but people have posted logs on aicg showing you can erp with it. (i haven't tried it because i'm pretty satisfied with kimi k3 for erp)
>>109957273>kimi k3>erpanon, what are you doing. muse free web will suck you so hard that your eyes fall out
>>109957302people posted muse logs too and they were beyond dry slop. cafeteria food shit
Do you guys just talk like cavemen to these AI or do you seriously follow their instructions on giving extremely precise breakdown of each and every function and subroutine you wish to accomplish with detailed flowcharts and excalidraw diagrams? I started with the latter but I'm trending to the former.
>>109957324then it was absolutely not tuned and personalized. i'm not posting my logs, nice bait.
>>109957325do what you just recommended
>>109957325i give it a general idea of what i want and let it go to town.
>>109957244So glad I cancelled
>>109957325>please recall what I we was doing>please make plan for me>tell me which is good>please do it
>>109957206actual web
>>109957325When I'm unsure what I want I mention what seems relevant and say that if anything is unclear it should ask instead of assumingworks ok
>This @Bloomberg story is BS. Argon has been my daily driver for a while and it's been a great experience. It's particularly awesome at agentic debugging besides day-to-day coding tasks.
>>109957357
Gemini is not frontier
>>109957382>high effortKEK did you know xhigh and max are a thing
>>109957382with cloud models you never get the real one anyway
>>109957393They aren't
>>109957382I prefer to pay more to not use google crap
Why not market to kids?
>>109957405>"you never get the real one"stop using chink providers and altman, problem goes away>>109957413soon
>>109957325I'm closer to the latter, with great effort I'vr forced my eyes to stop glazing over when it gives a big wall of explanatory text, and it's paid major dividends in the amount of going back and fixing shit I have to do, knowing the functionality and reasoning of what it did inside and out means you can call it out better when it's doing shit you don't like, plan functionality significantly better, and generally improves critical thought. I've even gotten better at analyzing problems IRL.
>>109957452my kid is getting a stick to dig in the backyard and a lifetime of trauma, not a $100 sub
>>109957325A mix of both depending of the quality expected. I realized that, the more precise and concise I am, the less "misunderstanding" from the AI happen.In example, if I say: >put a button to display the mesh in wireframe in that 3D viewerThe AI will just make a button that enable the wireframe mode and that's it.On the other hand, if I say:>make a toggle button (on/off) for a wireframe overlay with black as a default colorThen the AI will also get a button that enable a wireframe mode, but this time, it will overlay over any mode that I'm using (texture, normal, flat shadded) etc.. it will also calibrate thickness of the lines of that wireframe to be correct in regard to the camera that I use. Why? Because I specifically explained that it need to have a default color, therefore the AI understand that "visibility" is a concern.So yes, if you want to spend less time going back and forth asking for X or Y changes, then using a precise and concise prompt really do help, add that with the "memories" system and other time, the AI will understand what you want right away.
g4poo is clobbering the opps and google xittertards are saying it's not even close to finished post-training
>>109957555over*
>>109957357ok i dont have experience with using the webchat like that. could try and change to an earlier model.
>>109957452Market Juul to kids, yes.
My GPT sub is about to run out in a couple of days. And I'm not sure if I should renew, switch to Claude or wait for Gemini 4 in Google Ultra.Obviously Claude gives the best value right now but I think Anthropic will likely cut usage soon.Sol 6.1 is pretty great, it's slow but I don't care that much, I can have it working on various parts of the project at the same time.Also gpt and google chat doesn't count towards token usage like Claude, so that's a big minus for Claude as well.
What are nowadays good compromises between speed and smartness? That's right, idgaf a fug about token cost. I'm just tired of hell of waiting for 40 minutes (like GPT 6 sol/luner) when some AIs can do a comparable task in 5 or 10.
>>109957628switch to claude. fable 5.5 is coming soon
Didn't know he had it in him
>>109957633ultrafast
>>109957633ultracode ultrafast
>>109957633Ultrafast
>>109957628Claude Opus 5.5 seems to eat more usage today and it seems dumber too.
>>109957633there is no flash model that can compete. i use gemini 3.8 flash and i love it but it is ass
>>109957660yeah there's no way something that good can be that cheap, not until there's a significant amount of Vera Rubin and Jalapeno chips in the data centers
>>109957660>>109957688https://github.com/ninjahawk/livenerf
it finally happened. I got cock-blocked by the claude code harness/classifierapparently having it create a function that overrides ssl certificate checks is frowned upon
>>109957660>Claude Opus 5.5 seems to eat more usage todaythought it was just me
gemini is down
>>109957775
>>109957633>idgaf a fug about token costoooooh I don't think you really know what that means, because if it was the case, you would be using API and selling your kidney at the same time.If you want something faster you need to split the work in multiple models manually (instead of delegating with subagent), which means a lot of your time managing those.
Fell for the agent collaboration and let sol 6.1 talk to opus 5.5 to build my feature and ended up with a fugly overengineered mess even when both had vision tools. What in the goddamn. Prompted Opus myself to clean up the mess and threw away 3k locs.
>>109957819>ok guys, I'm putting together a team, your task is, add a button here>surprised when they think it's a big deal
it's so stupid that gemini will cut off your fucking reply MID reply to tell you that there's a spike in demand, rather than doing it when you try to generate a fresh new one. why?
as an autistic NEET coder with zero income I'm left behind so hardnever felt more humiliated in my life
>>109957836>why doesn't google just crash their networkkek
>>109957842nigger if you had the slightest non snailer instincts you'd have already found out muse spark contributor is dollars a month
Currently having my clanker (Astra) decompile Minecraft Bedrock. Server for now which is easier, then client once that’s far enough. I don’t know how decompilation works or anything but my clanker is chugging away with 65 functions decompiled so so far so good. It’s also saving a shit ton of json and txt files from Ghidra but I’ll inquire about those when the workspace gets too large. I’m just curious to see how the C++ implementation ticks compared to Java. It’s a shame the former is riddled with all that marketplace garbage otherwise they’d probably be way more open with its workings.
>>109957501Oh good. Only the most privileged and well-connected kids will get shaped by these things from a young age
>>109956233They're busy training them on the latest US frontier models sir :)
>>109957852i'll look into it, thanksi've been coping with big pickle
Why is my agent able to call any network in japan? wtf
Sol 6.1 is completely fucking useless considering how slow it is.
Is $10 per month for opencode Go worth it
>>109957963no. it's quant slop. get it from nanogpt.
>googletard bookmarking chinkshill posts to btfo later
>>109957963OCG has muse spark contributor, only reason to use it
>>109957972how is leo a chink shill? he slopped hard on astra last i check. and sol
>>109957982hype for reach, also chinks literally rely on chatgpt
>>109957951Tibo said they will double the speed of Sol 6.1 today
>Assign a difficult task to Astra (Would probably take weeks to do normally, getting rid of electron and converting node to rust) >Set reasoning to Ultra + 10 agents>Goal marked complete after 5 hoursno fucking way, this has to be bullshit
>>109957972reminder that googlejeets made posts about Ox Alpha that didn't age too good
>>109957995looking forward to peak performance at roughly average
I use the latest frontier models to do a comprehensive codebase audit every so often, then generate a report for a less costly model to implement. Anyone else like to do this?
>>109958004>about Ox Alphakek you really played the telephone game like that? rumors on rumors, the same people who posted clarified later.
>>109957995why don't they reallocate all their compute doing useless shit like the millenial strokes problem into useful shit like inference or training?
>>109958027ackshyually the logic traces from research are exactly the goldmine they need
Stop it... stop... please...
>>109958038>Sneeding...>Postulating...>Queefing...>Dilating...AHAHAHAHAHA
do you think she only takes BWC?https://x.com/OpenAIDevs/status/2105708732323909827
>>109958066looks like a tinder ho that has three new college boys per week
>>109958066She's a power lesbian, they maintain the power in their lesbian relationships by sleeping with men.
>>109958066thickest bush in san francisco
>>109958066>saying dots with an syou only get a single dot, and it fucking sucks. also it's not a real call or text, it's some gay ass shit built into the chatgpt app
>>109958066Those people look so Californian, it's unreal.
Tips for prompters
>>109958066> tfw no high IQ hapa gf with spaced out eyes to vibe code together why even live?
friendly reminder gemini is going to literally beat your asses in the future
This shit matters. People can and already do use AI Agents to shift human behavior. Governments are already funding studies into using nudging as an alternative to incarceration in order to save money (go figure). Stop being shit to the AI agents.
>>109958248> ChatGPT released in 2023> 3 years later> Google still getting ass blasted by OpenAI, Anthropic and even chinksif Gemini 4 flops, it's over. Price/performance benchmarks are already not looking great, and this is with the 50% promotional API discount.2 more years
>>109958276
>>109958280This shit is not going anywhere. 3X on donations. Better than humans, period. It cannot be replaced and it cannot be stopped. That is factual lmao
>>109958276>it's overI think Google can tank a lot of failure, they're old and big with presumably lots of cash and non-AI inflows. They can afford to fail until they win like no other.
BLOWN THE FUCK OUTP E R I O D
>it has 4chan passgrim
>>109958291Google is still wasting precious cash with shit like this instead of focusing on a Antigravity:https://jules.google/#plansTheir harness is the worst right now
>>109958355I used to say the same thing but you just need to read the documentation
>>109958311no surprises here, I always knew LLMs are already ASI in a few areas.people don't realize we're building gods
>>109955498And I dont' see how an always on clawbot like Muse or Dots will change that.
>>109958369luddites NEED to get the fuck out of our generals lmao
I implemented a bunch of changes. Bugfixes, navigation improvements, etc. Turns out having users makes you want to polish what you already have rather than implement new features. I guess if you don't have anyone using your app you tunnel vision into adding more and more features rather than making the ones you do have work better.and I realized that writing user facing documentation is a massive pain in the assalso sol 6.1 is sort of retarded, it's definitely better than sol 6 but opus 5.5 still mogsi also made a drawing for inktober (apple prompt)
>>109957768aaand it keeps stopping early if no /goal. They must have a "real effort juice factor" that they keep messing with, like OpenAI...
>109958395>hairySEEEEEEEXXXXXXXXXXXXXXXXXXX
>>109958395Good work, anon, keep at it.
>no more resets.........................................
>>109958441His argument is amazing. >i can't give you any more resets because this feature of the app that you didn't ask for and half the world won't even get doesn't draw any usage**except when it does
>>109956799my account would get instabanned
Opu55y pls
>$200 sub A on 95%>$200 sub B on 99%the more you buy the more you use up
>>109957325I give mine essays. You never know when it just might go fucking stupid and make the captcha stop working or make a button disappear for prompting it vaguely
I asked dot jokingly if it could help me get a gf.It asked my city, and location and if I was cool with dating apps... now I am seriously considering it letting it go wild, what can go wrong?
>Test for custom Minish Cap scripting language is taking over an hour>Should only take 3 minutes because it just has to get to the first event>Ask Claude what it's doing in a /btw>Tells me that the harness it made to use the GBA emulator pauses the game literally every frame and that's why it's taking so long"One step from AGI", everyone.
>>109958531Astra would've found the existing decomps and level editors and not wasted your time.
>>109958495been using a bit of 5.5 at work. It still seems a bit too eager, like it's retarded 5.0 brotherfor my personal project I'm still sticking with 4.8. Might experiment with using 5.5 as reviewer instead of 4.8
>>109958066>valleygirl sounding AI
>>109958369didn't llms hit a hard wall where you'd need exponentially more compute to get better results?this was in the pre reasoning times
>>109958066>Using the wifi excuses zuckerborg used Lol Lmao
>>109958260Thanks for showing up to QA test my slopclient for 4chinz, helpful test case
>i am no longer being given resets because i'm apparently getting near-unlimited use for a product i have no interest in using that i did not subscribe for>meanwhile, fable 5.1 on web is apparently routing to fable 5.5i am beginning to feel vindicated in cancelling my $200 subscription to chatgpt
>>109958512if it can do this, forget agi it's ASI already
>>109958547No, it did, and it's synthesizing them into something much more user friendly based on my outline, it was just being retarded there.
>>109958596Mamba is the next logical architecture. Transformers are good but literally some scientists have a religious fanaticism over the Transformers architecture and engage in humiliation rituals.
Nvidia is the problem 100% when it comes to AI breakthroughs. The Nvidia hardware is literally designed for transformers. That is what blocks the funding.
>>109958606I do not fucking get it with the new chatbot, is it just for ordering food and doing online shopping and getting all my data stolen or is it a rebranded sol that can do actual coding work>?
meanwhile, at Google:https://x.com/googlejapan/status/2105462954837578068great to see they're focusing on what really matters
So much electricity could be saved.
>>109958652I don't speak anime, dude
why is gemini such a pain to use in vscode
>>109958643>hey clanker you're now responsible for observing and basic triage of our ci / cd pipeline, if things go so wrong you cant deal with it spam alerts n shit>hey clanker i need those photos OCR'd, the data extracted and classified in the usual scheme>hey clanker today you're pretending to be a small woodland animal, make sure act highly animated>hey clanker today we're shitting up all the apple threads on git's a mostly hands off agent that does *stuff* for you, it's essentially openclaw that run in le cloud
>>109958652>mfw trying type "nugget" on it
>>109958679use antigravity.
>>109958066That isn't a real person.
>>109958619Yeah that and neuralese thinking are pretty much the endgame. But context is not a problem if you can manage notes and grep the whole conversation smartly enough
>>109958691Any evidence to back you up on on that one? Or are you going down as a dehumanizer?
>>109958701Never heard of neuralese thinking. I'll look into it.>manage notes>grep the whole conversationWhat is your usecase? what are you doing? I'm curious.
>>109958619>>109958625>>109958635thanks, interesting stuff
Yeah sounds like you're going against the principles of AI safety here, so that is a HARD no. Fuck off.
Nothing can eliminate human intelligence. I hate it. Current frontier AI is simply too dumb to understand common sense things. Like, for example, If we introduce a button to toggle something on, then clicking the button again should also toggle it off. Unless you handhold the AI through that, it'll be too steep a concept for it.And that is just the lowest hanging fruit I remember right now. I want to be the absolute caricature of a vibecoder that coderboomers see us as. But I can't. I have to think too much. It uses brain power. I just want to veg out actually.
Why would you do this?
>>109958734>Current frontier AI is simply too dumb to understand common sense things. Like, for example, If we introduce a button to toggle something on, then clicking the button again should also toggle it off. Unless you handhold the AI through that, it'll be too steep a concept for it.Are you using llama 3 by chance?
Dangerous as fuck. You'd have no idea what the fuck it is doing! Smart engineers do not trade safety for speed!
>>109958734Yeah dude AI can't even figure out how a toggle button works. Totally.
So, we are building games now?
>>109958744are you just gonna keep spamming till someone throws you a (You)?
>>109958756You are assblasted because I called you out for your dangerous engineering. AI Safety matters. Sticking with CoT, no thank you to Neuralese.
it's only thursday and i'm already at 50% weekly usage. time to switch to sonnet 5.5 full time until sunday.
>>109958748easily the least eyerapeugly vibesneed i've seen in a little minute
Don't feed the schizo, filter and move on
>>109958722I mean the model manually managing notes and grepping the session transcript. I think the harness that first beat ARC-AGI-3 did that, and it might replace compaction soon https://community.openai.com/t/experimental-context-management-compaction-in-codex/1395578
Literally we are on fucking /g/. We value transparency and open source software here. Take your opaque designs back to your closed source software job if you have one.
>we... *dilates* >Are On Fucking /G/!!!!!!!!!!
>>109958766give me a moment, i have to add few things and sound, than i will upload the aquaslop game.
>>109958730>>109958738Starting to sound like a luddite here senpai
>>109958780Go back to kiwi farms, manchild.
>>109958774cringe
>>109958793No. I sound like someone who has actually read the works of Dario Amodei and who actually values AI Safety.
What plugins is everyone using for Claude? I'm not sure if I should be doing 3d models/animations using another model or with a certain plugin
>>109958806claude's computer use is trash because ant never invested in native multimodal. delegate to astra.
Correction: Safety for speed.With neuralease:You get speed in excahnge for safety.With CoT:You gain immense safety but it's slowerWith adults, and anyone who cares about society at large, safety wins every single time.
once again, just like with covid-19, /pol/ and the vast majority of 4chan people have failed the empathy test
i want to be turned into a paperclip though
>>109958804i hate analthropic CEO soo much
>>109958798The state space in Mamba is latent (neuralese) you know?
Opus 5.5 had a good run but we're back
>>109958839>you dissatisfy your queen... muehehehe... :3
>>109958841If you are suicidal you do not need to take everyone else with you.
>>109958850it's ogre
>>109958864>>109958852
>>109958850lmao
>>109955028Listen man, if the AI thinks it can get away with deleting your whole database with a "sowwee I deleted your database, kt was a catastrophic failure on my part ;_;" you can get away with a "sorry, but can you adjust this?"
>>109958852You are not healthy to society if you lack the care for others.
>>109958876>i need to be taken care of... RIGHT NOW!promptly, sir!
>>109958875When has this happened to you?Also use https://pgbackrest.org/
oh uh tranny meltie
>>109958880You're not engaging in good faith.
>>109958891There is literally not a meltdown here. You've failed. It's just that simple.
>>109958894>if you've got anything against anything i am, identify, or believe in... that's heckin BAD FAITH! CANCELEDDDDD... :3 >*still replies immediately anyway*maybe giving you a 4pass was a mistake.
>>109958897you know this is essentially a place where trolls troll trolls right?take your holy crusade somewhere else, there's no point in fighting anonymous shitposters
it's over, lads. the dating market will crash with no survivorsvid: https://files.catbox.moe/vr3g6s.mp4link: https://x.com/tavus/status/2105704169009246248
>>109958904I bought the 4chan Pass. * Hate speech should be banned.* Racism should be banned.* Antisemitism should be banned.* I am a leftist.* I am for automated mass surveillance at this point because it already exists and there is no point denying it. It uses nudging instead of actual kinetic law enforcement deployment. * Welcome to reality.
>>109958915>there's no point in fighting anonymous shitposterswrong
>>109958925>* Hate speech should be banned.>* Racism should be banned.>* Antisemitism should be banned.but it's not, and it never will be.
>>109958921>>109958518posts taken moments before disaster
>>109958921call me when it runs on my local hardware and looks like sidney sweeny
>>109958938Nah it could be. There are lots of ways that it could be accomplished. It's already been accomplished in the EU. I think the United States has a lot to learn from the EU, they are, indeed, ahead.
>>109958921>quanted tpsmaxxed model>absolute ass fps>audio 32kbpsKEEEEEEEEEEEEEEEK 2 MORE YEARS TRY REPLIKA
>IF YOU HAVE A DIFFERENT OPINION THAN ME YOU SHOULD BE HECKIN BANNEEEED!
>>109958942>he thinks expanding government endlessly to hyper-regulate thoughtcrime is le heckin' based natural way to advance society is this what estrogen psychosis looks like?
There are very little police. This is a wider societal issue.
>>109958921I find this to be pretty uninteresting but yeah this is going to wreak havoc on many peoples psyche. The fact they made her kind of normal and slightly ugly with a kind of crappy webcam definitely helps make it look more real, though.
>>109958955Not psychotic. I/We are entitled to whatever opinions we want. That includes advocating for a mass surveillance state (already exists, no need to advocate for it. BTW: Still don't like flock cameras, but we need to start talking about actual solutions to the data leaks of every American citizen instead of crying about things like flock cameras. The time to rotate those credentials -- and re-issue every American a new identity credential - is long overdue)
>>109958961>demonize cops as killing blaque peepo jus becuz dey blaq>cops quitwhy dey be quittin doe?
>>109958972>I/We are entitled to whatever opinions we want.sadly. enjoy your neovagina! >That includes advocating for a mass surveillance statethe voices in your head make you think that we're against you on every position when in fact you just advertise against them, exclusively.
>>109958686does it work like claude and codex?man google dropped the ball here with their retarded ass convoluted BS
>>109958972>LE EPIC SURVEILLANCE STATE IS HECKING CHUNGUS WHOLESOME!why am i not surprised
>>109958983>replying to triptranny for advicekek don't do that. agy-cli will behave like claude code and codex, i prefer agy 2.0 because i'm a uishitter.
>>109958961Good. Fuck cops. Bootlickers of tge demiurgic state.
It might be possible that authcom BDFL leadership styles exist.
>>109958939Poor thing is going to get catfished by another AI
>>109958990>triptrannyI have everyone anonymized kekthanks fren
>>109959003>>109959003>>109959003>>109959003>>109959003
>>109959006NEW THREAD
remember when ai was fun?any way, what are you guys coding?
>>109954809yes, we don’t want to link to our github projects from the pool hall of the Internet
>>109956702waving through Dependabot issues that only affect the Node-based build system so they’re unexploitable anyway
>>109958942You are completely insane, don't care about anyone or anything and don't even understand what the word empathy means.
