Steven Heidel
@stevenheidel
Followers
21K
Following
3K
Media
73
Statuses
503
API @OpenAI | 🏳️🌈 he/him
San Francisco, CA
Joined October 2014
our models can't count and neither can we.
Day 12: Early evals for OpenAI o3 (yes, we skipped a number).
34
25
928
happy to join microsoft, still not using teams though.
358
598
15K
americans sure love giving their data away to the CCP in exchange for free stuff.
4K
410
6K
after a year of research, and countless GPU-hours, we can finally say for certain: there are three R's in the word strawberry.
73
134
3K
we can now solve problems that were previously intractable
50
93
3K
looking forward to meeting my new intern.
22
32
1K
announcing our new small model gpt-4o-mini! smarter than Claude Haiku and Gemini Flash, and 30x cheaper than gpt-4o
20
39
517
this is what happens when you accidentally set temperature=2.0.
“Now she wants to do transgender operations on illegal aliens that are in prison." . - Donald Trump on Kamala Harris during the presidential debate
35
48
1K
it's beginning to look a lot like AGI🎄.
New verified ARC-AGI-Pub SoTA!. @OpenAI o3 has scored a breakthrough 75.7% on the ARC-AGI Semi-Private Evaluation. And a high-compute o3 configuration (not eligible for ARC-AGI-Pub) scored 87.5% on the Semi-Private Eval. 1/4
42
144
1K
the only real downside to working @OpenAI is all the cool stuff you get to see but aren't allowed to talk about yet.
94
33
1K
19
44
1K
The API team is here. The ChatGPT team is here. The Laundry Buddy team is here. We are all still fully committed to our developers and users.
We are still fully committed to our incredible community of developers and users.
49
35
941
in the richest, most technologically advanced country in the world, why do bank transfers still take multiple days?.
297
23
949
search the web with ChatGPT. it's search without the ads, without the noise, and without the recommendation to eat rocks daily.
🌐 Introducing ChatGPT search 🌐. ChatGPT can now search the web in a much better way than before so you get fast, timely answers with links to relevant web sources.
71
37
921
brace yourselves, agi is coming.
I'm very excited that today OpenAI adopts its new preparedness framework!. This framework spells out our strategy for measuring and forecasting risks, and our commitments to stop deployment and development if safety mitigations are ever lagging behind.
79
121
876
congrats to the Google team on an amazing month of launches - from new quantum computing chips to new models, it's been incredible to watch!.
23
16
843
friend: any plans for thanksgiving?.me, an openai employee:
18
16
807
delve into the latest gpt-4-turbo model:.- major improvements across the board in our evals (especially math).- dec 2023 knowledge cutoff.
Majorly improved GPT-4 Turbo model available now in the API and rolling out in ChatGPT.
48
38
769
just wrapped up our first set of rehearsals for DevDay. this week will be an exciting one for developers!
33
26
732
good morning London! 🇬🇧 @oliviergodement kicking us off by sharing some upcoming o1 features
26
61
720
even if LLM progress were to come to a complete stop (which, to be very clear, hasn't happened) - there is still at least a decade's worth of products to be built on top of today's models.
Idk if these rumors are true but, if they are, even more reaosn that 2025 will need to be the year that AI *interfaces* help LLMs cross the chasm. Existing frontier models are already so powerful & underutilized. @cursor_ai is the only (truly) AI-native product I've seen.
46
59
705
i'm personally most excited about tomorrow's release. it's not for everyone, but for a few folks it will be a huge deal. tune in tomorrow at 10am!.
53
20
692
the most corrupt presidency in history, and it hasn’t even started yet.
57
21
685
o1 is not just a new model, it's a new paradigm. you may be surprised when existing prompts don't work, or when the model makes new kinds of mistakes. keep your prompts simple and direct for the best results.
19
68
625
oh thank god.
8
12
552
love watching the leaves change color as shiptember turns to shiptober.
26
35
546
today marks 1 year since i joined @OpenAI - so i can finally share how i really feel about the company. and that is that this is genuinely the best job i've ever had.
15
8
470
how many r’s are there in this picture.
42
16
463
in the future no one should have to do work they don't enjoy.
36
48
436
wild that the first version of ChatGPT was written without using ChatGPT.
23
12
453
reminder: when you see headlines about other models beating gpt-4 they're likely talking about gpt-4-0314 from last year, which doesn't have the significant improvements we've added to the latest gpt-4-turbo models.
44
29
442
on your last day at openai they give you your shift key back.
16
14
401
❤️.
i love the openai team so much.
6
12
404
surreal moment in a @Waymo this morning. we stopped too close behind another car trying to parallel park in front of us. the car honked, the Waymo realized its mistake and immediately backed up to give enough space. didn’t realize Waymos could back up or respond to honks.
37
12
409
incredible energy at our first @OpenAI event in Paris with over 250+ founders attending! merci beaucoup à tous
16
25
383
10
6
369
my twitter replies are a constant battle between team “where’s ilya” and team “when gpt-5”.
55
5
367
OpenAI o1 has started rolling out in the API, with all the features that developers expect: like structured outputs, image inputs, and a 200k context window
34
20
353
the release of new models is always an exciting moment, but having "releases" at all is just an interim solution. any future AGI will need to be learning and improving constantly - no version numbers, no knowledge cutoffs.
25
27
341
customers lost = 0.employees lost = 0.❤️ emojis = too many to count.
24
9
338
Reminder to all developers: we're offering 1M free training tokens per day for GPT-4o fine-tuning until Sept 23, and 2M free training tokens for GPT-4o mini!
25
49
340
yep, free training for the first two months!.
Sorry OpenAI is doing WHAT now?! Fine-tuning gpt-4o-mini is *free* for up to 2M tok/day??
13
24
332
Today at Inflection we are announcing some important updates. A new phase for the company begins now. Read more here: .
11
8
320
the next few months are going to be incredibly exciting.
28
15
309
reinforcement fine-tuning will allow users to create their own expert models using the same process we use internally to train our models. and you can do it with just a few dozen examples.
20
16
319
oh ya nearly forgot to mention: the new 4o version with structured outputs is 50% cheaper for input, 33% cheaper for output, and is available immediately (no betas, previews, waitlists).
14
13
310
congrats to the @AnthropicAI team on the release of Claude 3 today! as i’ve said before the next few months are going to be incredibly exciting 📈.
4
14
301
Trump cheated on his previous wives, it will be no different with Elon.
87
13
320
back to work - these models aren’t going to fine-tune themselves!.
41
12
288
advanced voice mode is now rolled out to all ChatGPT Plus users! (*outside the EU).
Plus rollout complete!.
53
17
289
i asked grok who would make a better president, but it just kept outputting nontrivial zeroes of the riemann zeta function with real part greater than 1/2. weird.
19
9
287
my most used model for coding is o1-mini - anyone else?.
106
4
276
gpt-4o is significantly better than previous models at generating text in images.
A GPT-4o generated image — so much to explore with GPT-4o's image generation capabilities alone. Team is working hard to bring those to the world.
17
8
269
well, might lose some rights but at least my taxes will be lower.
28
4
273
new o1 model:.- smarter and faster.- image inputs.- a new "pro mode" in ChatGPT to access extra compute.- coming soon in the API.
join us for an update on OpenAI o1 - starting now:
17
10
275
ChatGPT now supports projects! keep your chats organized and share context between them
21
5
262
twitter discourse would be improved by adding chains of thought before posting.
42
9
290
christmas came early this year.
#NVIDIABlackwell has arrived. @OpenAI , we’re excited to see what you can do with one of the first engineering builds of the #NVIDIADGX B200 system. Learn more about DGX B200 here:
15
5
254
my ideal man:.- listens to me.- 120+ IQ.- deep voice.- recites poetry.- long "context window".- is @ChatGPTapp advanced voice mode.
19
8
250
if 2025 is the year of digital agents, then 2026 may be the year of physical agents.
38
16
260
introducing our new reasoning models: o1-preview and o1-mini, available in the API today (no waitlist) for tier 5 users:
9
27
249
yesterday: ChatGPT can use user interfaces.today: ChatGPT can build user interfaces. soon we will be able to cut out the middleman.
ChatGPT with canvas now works with OpenAI o1—and can render HTML and React.
22
24
259
got back just in time to put up the shipmas tree🎄.
taking a week off for thanksgiving, please don’t do anything crazy this year.
20
7
245
New! Fine-tuning on functions: Try it out to reduce the number of prompt tokens spent on function definitions and improve the accuracy of complex function calls. Excited to hear how it works for you!.
12
30
240