Latest sick burn just dropped
>>108870357Aha, what a clever and humorous twitter post! Thank you for sharing, OP.
>>108870357>4oOAI and Anthropic's internal models are very close to AGI, these old public models mean nothing.
>>108870401Holy mother of cope!
>>108870357The image shows a screenshot of a conversation with an AI model that incorrectly identifies 9.11 as larger than 9.9. The model appears to be treating the numbers like software version numbers (where v9.11 would indeed come after v9.9) rather than decimal values. To compare these two decimals, it helps to give them the same number of decimal places: - 9.9 is equivalent to 9.90 - 9.11 is 9.11 Since 9.90 > 9.11, 9.9 is the larger number. The actual difference between the two numbers is: 9.9 - 9.11 = 0.79
>>108870357>sikh bern
>>108870401Every new model they release they say it's too dangerous to release and could be AGI. For years they've been doing this grift and you're still falling for it.
>>108871354only gay faggots like claudropic are doing that
>>108870401>two more weeks
>>108871354except it actually solved a well-known unsolved problem and everyone using them daily knows theyve gotten exponentially better over the past yearluddites are huffing that final bottle of copium before they drift into the endless nightmare of permanent unemployment in hypercapitalist muttland
>>108870357lying piece of shit
>>108871412>several years of grifting, trillions of dollars of investment, billions of dollars on electric costs, hundreds of billions on hardware>solves ONE math problem that has no application and was a curiosity to some mathematiciansokay let me know when they can do something useful like write working software or answer basic fact questions without getting 70% of it wrong
>>108870527>saving this image 10 years ago with that filename just to use it on the off chance someone says "sick burn" in a context that involves indians kek nice
>>108871412>20 years just to make a very tiny progressyeah AI is totally advanced
>>108870357>>108871475Luddites have to resort to using old outdated models like 4o to try and make their points seem more valid. Pathetic, really
>>108871482what model and version specifically did you last use that was unable to do those things and how many years ago was it. assuming you have ever used an llm a single time
>>10887035750% of people probably can't figure out if 9.11 is bigger than 9.9 so the fancy autocomplete is dumber than 50% of people. is that actually a burn?
for me, 3.5 flash gave wrong answer and 3.1 pro gave right answer
You haven't been able to talk to 4o in months.That meme of >used chatgpt once in 2023Is very real
>>108871585my little pepe?
>>108871597shhhhhhhh dont alert the janny
Just tell me who luddite chuds voted for so i know which side to hate.
>>108870401(you)
>>108871475>>108871515inb4 it is a literal in the base prompt to prevent the mistakeTry some different values, preferably in new contexts.
>>108871515>>108871475>ai dev browses /g/>sees people exploit a limitation in a model>reinforce "correct" response weights by adding even more preemptive context into the generated prompt, and passing the cost of extra compute to the customers
>>108872211>>108872213The solution is much easier and it will work with any values. For math Claude and Chatgpt run a self-written python script, then visualize its outputs and solutions. LLM will always get math wrong, but new models know how to leverage other tools to get the correct answer
>>108872258But if you don't put the values into the python script correctly, you'll still get the wrong result.
>>108872271Yes. But it's pretty good at putting values into the script it writes itself. You might get false results if you try hard enough. But in vast majority of cases it will be able to put values into a script and show you the result
