Eric
970 posts
Eric
@ericmitchellai
chatgpt posttraining @openai. building personal agi. I like ai and music and some other stuff
United States Katılım Aralık 2017
583 Takip Edilen10.9K Takipçiler
@charliermarsh @andrew_n_carr wow, sick, welcome Charlie! Go tigers!
English
We've entered into an agreement to join OpenAI as part of the Codex team.
I'm incredibly proud of the work we've done so far, incredibly grateful to everyone that's supported us, and incredibly excited to keep building tools that make programming feel different.
English
@ericmitchellai @atulit_gaur Can chat not speak in bullet points for pages. Thanks :)
English
@ericmitchellai @atulit_gaur Hi eric, catastrophic things happen when I ask GPT-5.4 Pro to generate an image - it spams the imagen tool and I end up hitting the rate limit
English
@theo @atulit_gaur thanks theo, glad you really liked 5.4. Really proud of the team for that one
English
@DavidOndrej1 Smoking incredible code
I would walk a mile for code from GPT-5.4
English
GPT 5.4 *is not* better than Opus 4.6
i have no idea what people are smoking
English
@ericmitchellai @nicdunz 5.4 constantly does this with me. It seems like using the web tool makes it behave like the prior message was the first one in the conversation. I therefore explicitly ask it not to use the web tool as it will mess up your work. It also makes it reasoning about the task.
English
@Miles_Brundage thanks miles. If you ever run into an example you feel comfortable sharing, would love to see <3 glad you're finding the model useful!
English
@ericmitchellai I don't love sharing links / logs since they often involve non-public docs but here's an example prompt -
"Given what you know of me, what should I read right now?"
It then suggested reading docs *from* the (ChatGPT) Project folder itself, not public reading material.
English
The most common failure mode I've observed with GPT-5.4 is misunderstanding the intent behind the prompt (but then doing a good job at what it thought the task was).
I'm not sure if this is a regression or not, but it stands out by contrast w/ the task execution
English
GPT-5.4-Thinking has a very non-annoying writing style which is nice
English
@_simonsmith @dejavucoder Thanks a lot Simon. Verbosity in 5.4 is actually on our radar already :)
Appreciate all your feedback!
English
I made a nonfiction writing benchmark to evaluate this and, as much as I love the writing improvements in GPT-5.4, it confirmed my observation that by default GPT-5.4 is excessively verbose. I have tamed it with custom instructions but here's a heat map showing its strengths and weaknesses relative to Opus and Sonnet.
English
@z @ericmitchellai I think this is just a bug. It gets confused what it is doing.
English
Chat GPT 5.4 is really frustrating to work with. Awful hallucinations and laziness. It's hard to tell if it's "smarter" when it's so difficult to steer.
English
