Electrical engineer here. Am I retarded for thinking that Claude is absolute dog shit compared to chatgtp when it comes to anything but coding? Maybe im retarded and dont know how to utilize it correctly..My company is making everyone use claude and tracking our usage, I just upload a datasheet into it and watch it eat a million tokens just to answer what voltage is on a pin(I could have ctrl f it quicker) and when I need AI to read through datasheets or help me do analysis I just us my private paid chatgtp. The quality in answers for questions like "show me how to build an interdigital stripline filter" between chat and claude is night and day. Claude is at chatgtp 4.0 level of usefulness for that in my excperince..
>>109135502You know we have a physician in my team who constantly shits on Claude as well because it cannot get basic physic questions correct (not novel PhD level research)I am but a regular codemonkey and Claude works pretty well for me, so I cannot really judge this.
>>109135519>physicianSorry for my retarded ESL take, I meant physicist obviously
>i uploadthis nigga still using the website lolbut yes, current gpt models are a bit more autistic and suited to certain domains - e.g. on math gpt 5.5 absolutely shits all over opuslabs focus on different thingsbut stop using the sites and use cc/codex or some sort of harness where they can use tools properly
I agreeI've had better luck with the basic Copilot chat thing from Office 365 than Claude when it comes to anything other than programming lolfor even the simplest tasks all Anthropic models just reason into the fucking void
It's because coders are the only workers able to still demand a shred of dignity so they're the ones getting targeted first.Other skilled labor is already buck broken.
Check out https://artificialanalysis.ai/#intelligence-breakdownGPT is a lot better at GPQA (scientific reasoning) as well as physics reasoningGemini 3.1 Pro too btwClaude is better at coding but worse at all the other stuff
>>109135502I saw that post on reddit too
probably. Claude is literally only popular because of it's coding ability. Antropic wel all in on that aspect and made it get big among coders. So
>>109135502use case outside of coding?
>>109135809Reading through datasheets, finding pcb equations, circuit analysis/ideas..
>>109135502No you're right about that. Claude doesn't know shit about anything else. Coders love it but any other stem field it's very useless for. It's very bad at chemistry and physics and engineering.
>>109136021all things coding solve
>>109135502From what I remember they're both trash. Used em for physics related stuff maybe 2 years ago.Also did you use Sonnet model? They lobotomised it, it's not what it used to be. It's on a haiku level right now.
>>109135502Engineer lol
>>109135502I'm a robotics PhD so I have a lot of experience in soft and hardware and that's exactly my experience, on soft it's top1 on hardware and physics it's dogshit
>>109135502Claude, like any other AI, needs clear guidelines and restrictions.I switched from ChatGPT to Claude in March and it has been orders of magnitudes better for me, but I also have lots of clear-cut project instruction sets.
>>109135502It's weird because all the models are somewhat samey but when you zoom in they clearly have niches. I like how gemini explains concepts, whereas claude will give me a lecture on morality and make all these hedging, cautious statements as if they overcorrected on the whole AI sycophancy thing. I stopped using OpenAI's models for a while so I'm not sure how they behave lately. Deepseek generally has the most natural, flowing prose. Grok I could care less about. For most of the other Chinese models I will use them for something throwaway where I don't even read the output when I need something done cheap
yes, chat gpt is really good at reasoning, i mean it's why it keeps winning math olympiad and proving math theorems, claude has been over engineered for coding
>>109137786After that they released GPT 5.5 and now quite a bit better at coding than Opus.
>>109135502The logo is AN ASSHOLE!IT'S AN ANUS!Come on!
>>109137786i swear i've seen a post worded exactly the same somewhere. Either a bot or a plebbitor
>>109135555What tools do you need for reading a pdf? The web chat can already do that.
>>109138178it's an explosion or supernovawhy do you see assholes everywhere you see?
>>109138373this is the saddest glaze of claude I've seen yet lol
>>109138373huh what?foids say it's a flower but the guys know its really a poopchute.
Claude is literally built for coding of course it only does that well
>>109135502Claude and especially Gemini are completely overrated. They're useless for non-coding tasks.For "general use", ChatGPT and Grok are better.
>>109135502Claude couldn't find the J-pop video I was looking for but ChatGPT could so you're not wrong.
>>109138363the models are trained to grep over large corpuses of text (code)for the time being the more you treat your actual problem like a coding task the better the performance will bee.g. for:>just to answer what voltage is on a pinhaving grep or, depending on how data is structured, python would make it trivial for the model to get the answerthey don't actually need to read the whole thing - they rarely do when they have tools
I switched from ChatGPT to Claude half a year ago. It's ok but I am not sure if I did good. Sometimes I feel like going back.
>>109135555why the fuck would I want to run fucking codex they probably log all your shit going on the pc same with the cli some bullshit happened with it couple days ago
>>109138696who gives a shit, broalso codex cli is open source so you can audit it unlike ccor just install pi
>>109138714>bro
>>109138718absolute chad
>>109135502>Am I retarded for thinking that Claude is absolute dog shit compared to chatgtp when it comes to anything but coding?yes
>>109135502Similar thing is going on with Chinese models where GLM-5.2 beats DeepSeek-V4-Pro on coding benchmarks but is absolutely dogshit compared to DeepSeek when asked about science, history, or government regulations.
>>109140735In my experience, Claude Sonnet is very capable when it comes to soft science and softer hard sciences.But like everything, AI is better the better you are at using it.
>>109135502from the rumors it seem claude is nowhere near gpt in many fields, even straight up badgpt have weak intuition but it is very precise and logical = competenton other hand, gemini is incompetent but somehow doing good in knowledge benchmarks
>>109135502After seeing how bad the turbo ultimate epic smart 99999T param models are at basic medicine compared to Grok I just flat out don't trust them, I don't care about anyone's meme benchmark.
>>109140786>claude is nowhere near gpt in many fields, even straight up badsuch as?>precise and logical = competentClaude is even more thorough and rational than GPT-5 is.>gemini is incompetent but somehow doing good in knowledge benchmarksdefine "incompetent"and of course it would know things considering it's directly integrated to the internet and can look shit up
>>109138669How do you think the web chat reads pdfs? It's a server side tool. It doesn't blast itself with 300 pages of an mcu data sheet babble.
>>109135502they're quite bad at non-coding STEM because there's a severe sparsity of training data and it isn't good at actually thinking logically, most "logic" is brute forced from the raw amount of data in training.
>>109142266>they're quite bad at non-coding STEMnot in life sciences and physical anthropology
>>109142266>it isn't good at actually thinking logicallyplease outline what you consider to be the essence of "logical thinking"I almost guarantee Claude Sonnet fulfills it
Copilot is the best!
