Apparently chat GPT 5 is underwhelming lmao
- Thread starter Scustin Bieburr
- Start date
chatgpt is orwellian and psycho.
the more they engineer it for generality the worse it gets with factual hallucinations.
Why is AI halllucinating more frequently, and how can we stop it?
OpenAI's most advanced reasoning model is smarter than ever — but it hallucinates more than previous models, too.www.livescience.com
they claim otherwise
The more advanced artificial intelligence (AI) gets, the more it "hallucinates" and provides incorrect and inaccurate information.
Research conducted by OpenAI found that its latest and most powerful reasoning models, o3 and o4-mini, hallucinated 33% and 48% of the time, respectively, when tested by OpenAI's PersonQA benchmark. That's more than double the rate of the older o1 model. While o3 delivers more accurate information than its predecessor, it appears to come at the cost of more inaccurate hallucinations.
1/11
@buildthatidea
openai just dropped gpt-5 and it's insane!
it’s smarter, cheaper and faster than previous sota coding models.
here's everything you need to know:
2/11
@buildthatidea
1/ gpt-5 is so much smarter than sonnet/ opus while being so much cheaper
it scores 74.9% on swe-bench, beating claude opus 4.1’s 74.5%.
3/11
@buildthatidea
2/ gpt-5 (with thinking) hallucinates less than openai's previous models:
- hallucinates 5x less than o3 on open-source prompts
- hallucinates less with reasoning on health bench
4/11
@buildthatidea
3/ gpt-5 pro scores less on humanity's last exam than grok 4 heavy
5/11
@buildthatidea
4/ gpt-5 gets a 95.2% score on the needle-in-haystack test
that’s a massive jump from gpt-4.1 at 57.2% and o3 at 55%.
it’s the best performance openai has ever published on long-context retrieval.
6/11
@buildthatidea
5/ gpt-5 achieves 100% on aime 2025
7/11
@buildthatidea
6/ gpt-5 is the best writing model openai has ever released
it handles rhythm, ambiguity, and form with depth and clarity
8/11
@buildthatidea
7/ gpt-5 can make web apps/ games in one shot
9/11
@buildthatidea
8/ gpt-5 has 4 new personalities
cynic, robot, listener and nerd
these are text-only, opt-in, and available in settings under customize chatgpt
10/11
@buildthatidea
9/ gpt-5 is a unified system that picks the right mode for the job
- fast mode handles everyday questions
- thinking mode kicks in for complex tasks
- a smart router chooses automatically based on your prompt
you can also type "think hard about this" to trigger deep reasoning
11/11
@buildthatidea
10/ want to launch your own ai agents with gpt-5?
sign up at BuildThatIdea - The world's biggest AI agent marketplace. it's coming soon
To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196
@buildthatidea
openai just dropped gpt-5 and it's insane!
it’s smarter, cheaper and faster than previous sota coding models.
here's everything you need to know:
2/11
@buildthatidea
1/ gpt-5 is so much smarter than sonnet/ opus while being so much cheaper
it scores 74.9% on swe-bench, beating claude opus 4.1’s 74.5%.
3/11
@buildthatidea
2/ gpt-5 (with thinking) hallucinates less than openai's previous models:
- hallucinates 5x less than o3 on open-source prompts
- hallucinates less with reasoning on health bench
4/11
@buildthatidea
3/ gpt-5 pro scores less on humanity's last exam than grok 4 heavy
5/11
@buildthatidea
4/ gpt-5 gets a 95.2% score on the needle-in-haystack test
that’s a massive jump from gpt-4.1 at 57.2% and o3 at 55%.
it’s the best performance openai has ever published on long-context retrieval.
6/11
@buildthatidea
5/ gpt-5 achieves 100% on aime 2025
7/11
@buildthatidea
6/ gpt-5 is the best writing model openai has ever released
it handles rhythm, ambiguity, and form with depth and clarity
8/11
@buildthatidea
7/ gpt-5 can make web apps/ games in one shot
9/11
@buildthatidea
8/ gpt-5 has 4 new personalities
cynic, robot, listener and nerd
these are text-only, opt-in, and available in settings under customize chatgpt
10/11
@buildthatidea
9/ gpt-5 is a unified system that picks the right mode for the job
- fast mode handles everyday questions
- thinking mode kicks in for complex tasks
- a smart router chooses automatically based on your prompt
you can also type "think hard about this" to trigger deep reasoning
11/11
@buildthatidea
10/ want to launch your own ai agents with gpt-5?
sign up at BuildThatIdea - The world's biggest AI agent marketplace. it's coming soon
To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196
