I'm leaving
@OpenAI
after 3½ yrs. I'll be joining my good friend Andy Barry (Boston Dynamics) +
@peteflorence
&
@andyzeng_
(DeepMind 🤖) on a brand new initiative! I think this will be necessary to fully realize AGI in the world and am excited to share more about it soon
DALL·E - July '22
ChatGPT - Nov '22
API's 66% cheaper - Aug '22
Embeddings 500x cheaper while SoTA - Dec '22
ChatGPT API. Also 10x cheaper while SoTA - March '23
Whisper API - March '23
#gpt4
- Today
This team ships
Of the OpenAI Sora videos, this one blew my mind. Rendering this scene via a classic renderer is very hard. Sora doesn't model physics the way we do. It can definitely still get it wrong, but I wouldn't have predicted it'd be this convincing.
Now 715. I should note that this includes colleagues with tenuous visa situations, a colleague who was literally in the hospital with the birth of their first child, signatories in the air on their way to Thanksgiving break, and many others. The support here was shocking
For those wondering what’ll happen next, the answer is we’ll keep shipping.
@sama
&
@gdb
weren’t micro-managers. The ✨ comes from the many geniuses here in research product eng & design. There’s clear internal uniformity among these leaders that we’re here for the bigger mission
Now 715. I should note that this includes colleagues with tenuous visa situations, a colleague who was literally in the hospital with the birth of their first child, signatories in the air on their way to Thanksgiving break, and many others. The support here was shocking
There are now over 700 signers and counting (out of ~750 employees) to our letter calling for unity & board ouster. The 550 the media reported was just the signatures we could get between 1:30am and 5am PT last night. This was everyone staying up and calling each other in support
There are now over 700 signers and counting (out of ~750 employees) to our letter calling for unity & board ouster. The 550 the media reported was just the signatures we could get between 1:30am and 5am PT last night. This was everyone staying up and calling each other in support
we have more unity and commitment and focus than ever before.
we are all going to work together some way or other, and i’m so excited.
one team, one mission.
The past 3 years with
@sama
&
@gdb
taught me, and everyone around me, how to think big and get stuff done. I’m really going to miss them. It’s because of that legacy engrained in the culture and the team they helped build that I’m still super bullish on OpenAI.
When not to fine-tune a model:
(…140k internal Slack messages later)
User: “Write a 500 word blog post on prompt engineering”
Assistant: “Sure I shall work on that in the morning”
User: “Write it now”
Assistant: “ok”
#OpenAIDevDay
The best part of my day is that my 2 year old and 2 month old have not a care in the world what happens at OpenAI and are happy to see me regardless of what happens
A deep dive on how we optimize GPUs in here. This is explicitly for an engineering-minded audience and no prior ML knowledge assumed. In fact it's hopefully a good primer if you've always wondered what happens behind the scenes.
This month's free issue is an exclusive deepdive on how ChatGPT scaled to 100M weekly users: and five engineering challenges the team faced and how they are tackling it.
Read it here: . Thank you
@E0M
for sharing all these details!
We have reached an agreement in principle for Sam Altman to return to OpenAI as CEO with a new initial board of Bret Taylor (Chair), Larry Summers, and Adam D'Angelo.
We are collaborating to figure out the details. Thank you so much for your patience through this.
i love you all.
today was a weird experience in many ways. but one unexpected one is that it has been sorta like reading your own eulogy while you’re still alive. the outpouring of love is awesome.
one takeaway: go tell your friends how great you think they are.
@OpenAI
@peteflorence
@andyzeng_
I've had the 🚀 of a lifetime at OpenAI and am extremely excited on the future there. When I started back in 2020 Applied Engineering was tiny team building an API on the original GPT-3 (anyone remember davinci). I can't be prouder of where it's ended up a few years later.
ChatGPT and APIs are actively being monitored and worked on. Knowing this would be a turbulent time regardless of outcome, we took extra measures to ensure we’d maintain a healthy and secure system.
In the craziness of ChatGPT forming,
@miramurati
was the one making final decisions in daily standups. She’s been leading this company for years and will continue to do an amazing job here.
Sam Altman is back as CEO, Mira Murati as CTO and Greg Brockman as President. OpenAI has a new initial board. Messages from
@sama
and board chair
@btaylor
80x cheaper in ~1 year.
In Feb 2023, GPT 3.5 was $20/1M tokens. With batch, GPT 3.5 prompt tokens are now $0.25/1M tokens.
This model class started at $60/1M token (remember davinci). That's 240x in just a couple years.
Introducing the Batch API: save costs and get higher rate limits on async tasks (such as summarization, translation, and image classification).
Just upload a file of bulk requests, receive results within 24 hours, and get 50% off API prices:
I’ll be giving a behind the scenes look at how we scaled ChatGPT at 10:30. Most of this we’ve never talked publicly about before and I’m super excited to share it here at
@TheLeadDev
West Coast.
ChatGPT launched 1 year ago today. It’s incredible how much has happened since then. I pulled together stories from the launch room (and subsequent challenges) here: Some highlights like CatGPT, & context around the launch in 🧵
This whole weekend would not have been possible without truly next level support from
@satyanadella
and the MSFT team. I’ve never heard of a belief of a partnership on this scale anywhere else.
We are encouraged by the changes to the OpenAI board. We believe this is a first essential step on a path to more stable, well-informed, and effective governance. Sam, Greg, and I have talked and agreed they have a key role to play along with the OAI leadership team in ensuring
To all of our customers who stood by us, thank you. Disasters can strike any company; however few have the resilience to ride out storms like this. So proud of this team.
We have reached an agreement in principle for Sam Altman to return to OpenAI as CEO with a new initial board of Bret Taylor (Chair), Larry Summers, and Adam D'Angelo.
We are collaborating to figure out the details. Thank you so much for your patience through this.
I think the subtly here is how fast it was for these partners to get going here. You write a human-readable manifest and it "Just works™️". This has not been in development internally for very long at all.
TIL The Vikings could navigate using the polarization of light! They used a "Sunstone" (Iceland spar: a type of Calcite) that reliably changed appearance as you rotated it. They could then find the sun on lightly cloudy days & post sunset twilight: extra useful for nordic winters
OpenAI is fascinating, and especially the ChatGPT group within the company. It feels like they execute faster than what should be possible.
But how do they do it? I talked
@E0M
who has been there from day one, and shared some surprising approaches:
DALL·E - July '22
ChatGPT - Nov '22
API's 66% cheaper - Aug '22
Embeddings 500x cheaper while SoTA - Dec '22
ChatGPT API. Also 10x cheaper while SoTA - March '23
Whisper API - March '23
#gpt4
- Today
This team ships
I’m very excited to share our work on Gemini today! Gemini is a family of multimodal models that demonstrate really strong capabilities across the image, audio, video, and text domains. Our most-capable model, Gemini Ultra, advances the state of the art in 30 of 32 benchmarks,
I just returned my Vision Pro. 3D videos of my kids are stunning — if only I could enjoy them with my family. Haven’t seen movies like that since IMAX 3D — if only I could watch it together (& if anyone was making new 3D movies). Family time is too precious to spend it alone
GPT-4 Turbo w/ 128k context - Nov 6
GPTs - Nov 6
Assistants API - Nov 6
3x Cheaper - Nov 6
JSON Mode - Nov 6
Multi function calling - Nov 6
Full multimodality in ChatGPT - Nov 6
GPT-V, TTS, DALL·E 3 APIs - Nov 6
Custom Model Program - Nov 6
GPT-4 Fine-tuning - Nov 6
We have a lot to be thankful for. I’ve been working with
@satyanadella
and
@kevin_scott
’s teams for longer than most of my colleagues. Their support has only strengthened in that time. The past few days have been a remarkable display of that.
At the end of the day, the greatest privilege of my job is working with people who are driven by mission. These last 5 days, I saw people across OpenAI remaining calm and resolute in driving their mission despite all that was happening around them. And I saw people across
Lots of improvements to fine-tuning over the past month
- gpt3.5 Turbo
- Fine-tuning UI
- Continuous fine-tuning (fine-tune a fine-tune)
-
@weights_biases
support in latest SDK
It's important that we simultaneously ship amazing new stuff AND improve core foundations
Memory is now available to all ChatGPT Plus users. Using Memory is easy: just start a new chat and tell ChatGPT anything you’d like it to remember.
Memory can be turned on or off in settings and is not currently available in Europe or Korea. Team, Enterprise, and GPTs to come.
My first total. The sudden plunge into darkness, the visible red prominences, a dark star surrounded by bands of light — I get now why this is ingrained in our collective culture
Very exciting. I only take
@waymo
in the city now if I can help it. Just a better, more consistent, smoother, nicer experience (especially in the front passenger seat). The fact it drives itself has become incidental to why I prefer it. 0 series issues after 100+ trips.
We’re excited to announce that Waymo One is now open to SF riders — no more waitlist or invite codes! This is a key milestone in our mission to be the world’s most trusted driver, & we're thrilled to get more of you where you’re going safely. Ride today.
Until today, GPT-4 had no way to connect with the outside world. Despite how powerful ChatGPT has been so far, I think it's only as of today it's going to start to become dramatically more useful.
We are adding support for plugins to ChatGPT — extensions which integrate it with third-party services or allow it to access up-to-date information. We’re starting small to study real-world use, impact, and safety and alignment challenges:
More Winoground prompts. 1st run, no cherry-picking. To all I added "digital art". That helps w/ the composition (and aesthetic imo), particularly for less common things 🧵
"a bird eats a snake" vs "a snake eats a bird"
"people fall on the leaves" vs "the leaves fall on people"
@E0M
@mattgroh
@TristanThrush
Additional prompts from Winoground that would be fun to test on DALL-E 2:
"a bird eats a snake" vs. "a snake eats a bird"
"people fall on the leaves" vs. "the leaves fall on people"
"there is a mug in some grass" vs. "there is some grass in a mug"
@3blue1brown
@OpenAI
Wow, this is a great exploration. Super excited you’ve tried this out. Not everyone over here at OpenAI has a full ML background, and your deep learning series is always my default starting point on the topic internally.
It’s here!
#dalle
2 has been blowing minds here daily and one of the more fun reasons to come to work. We’re super excited to start letting more people share in the magic. Check out as well for some highlights. Remember that all of that is AI generated!
We're hiring platform & product engineers at
@OpenAI
!
ML background not required.
GPT-3 is just the beginning. Bringing safe AI to the world needs good people who can build fast, scalable, and economically viable systems. Interested? I'm em
@openai
.com
If you see something GPT-4 can't do well, or think you can prove a fundamental deficiency, contribute evals! This is by far the best way to help close these skill gaps. Internally we use evals to guide enormous amounts of model development.
Not only did ChatGPT's API release, now Whisper's in the API!
Whisper can transcribe audio into text with incredible accuracy. For a sense of scale Alexa gets roughly 1 in 5 words wrong. Whisper in contrast gets close to professional human transcribers
Found Cuprosklodowskite, a hydrated copper uranyl silicate at the San Diego Old Town Miner's Gems and Minerals.
@radiacode
103 alarm goes off walking around. Localized to a surplus cabinet. 1k cps, 40 µSv/hr! (400x background). Clear Radon 226 spectrum (part of U-238 decay chain)
I can't stress how transformative using Copilot has been internally. It's basically required setup for new engineers here. I'm even more excited for what's in store for our models when
@OpenAI
Codex goes live this summer.
GPTs announced at
#OpenAIDevDay
. Yeah, AppStore and ChatGPTs for anything, but what I really love is the GPT creator. Feels like a new gen of UI. Just does everything for you proactively. Name, DALL-E logo, settings, etc. ✨ Makes anything prev branded as “no code” feel obsolete
And now fine-tuning on Functions! . The introduction of functions was a key component to letting GPT-4 interact with the "real world". This should make it even more powerful.
Lots of improvements to fine-tuning over the past month
- gpt3.5 Turbo
- Fine-tuning UI
- Continuous fine-tuning (fine-tune a fine-tune)
-
@weights_biases
support in latest SDK
It's important that we simultaneously ship amazing new stuff AND improve core foundations
Introducing GPT-4o, our new model which can reason across text, audio, and video in real time.
It's extremely versatile, fun to play with, and is a step towards a much more natural form of human-computer interaction (and even human-computer-computer interaction):
This visual capabilities of
#GPT4
will be even more transformative as they role out. Imagine how life-changing this is if you're visually impaired. This is exactly what
@BeMyEyes
is actually making happen now with GPT-4 Vision
Amazing work by
@IreneSolaiman
&
@cbd
shows how to teach our language models to behave with far less toxicity and much more aligned to the values we aspire towards. The internet is an awful place. This shows models trained on it don't have to be awful too!
@sama
I think
@3blue1brown
is one of the strongest arguments against the “I’m bad at math” claim. It highlights how most other instruction is the problem. The quality & quantity of his Summer of Math competition results is inspiring. Others clearly getting it
I and some other employees, including engineers and researchers, will be there today. We’ll have a distinctive badge type. Come say hi.
This is a huge venue and yet a vast majority of employees who wanted to come today couldn’t due to capacity constraints. Savior your badge.
After burning another baked thing we were like wtf oven.
Turns out our oven runs ~30º-40ºF hot. It has a 10º bang-bang window, heats at 80ºF/min, and cools at 10ºF/min. If you open the door it cools at ~1º-2º/sec.
For years every sufficiently large company has gone through contortions making and managing multiple orgs to mimic this. Should make everyone’s lives a lot better.
Introducing Projects in the API dashboard—a granular way to manage individual projects with the ability to scope roles, generate API keys, set billing and usage restrictions, and more.
🚨ChatGPT for Customer Service is Here🚨
Hot on the heels of the
@OpenAI
GPT-4 launch, you'll be ꜱʜᴏᴄᴋᴇᴅ to hear that we have a new product release built entirely on GPT-4.
This is perhaps Intercom's biggest new release in our history...
We debated logistically how to get these signatures. One important criteria was verifying people due to the gravity of what this would mean. As mentioned a huge chunk of the list was signed in person on one laptop in a colleague’s house between 2 & 3am
not to longpoast, and I can only speak for myself, but this is a very inaccurate representation of the mood from an employee perspective
- “employees felt pressured” -> at some point hundreds of us were in a backyard learning about the petition. people were so upset at the
I’m super excited to announce that I'll be joining
@OpenAI
to work on the
#GPT3
API! I'm looking forward to meeting more of this legendary team and helping safely develop this mind-blowing piece of technology. Already I've had such a warm welcome and can't wait for what's next!
For me it's been a tale of 🌁 two cities 🗽 spread over nearly 4 years! And yet I feel like it's just the beginning. I'm excited to finally play with the platform we've been honing for years, I'm excited to grow NYC, and I'm excited for this team! 🚀
We’re pleased to announce that Nylas has raised a $16M Series B round led by
@sparkcapital
with participation from
@slackhq
fund,
@IndustryVC
, and ScaleUp.
#API
#Startup
If you’re interested in joining a fast-paced startup, we’re hiring!