chatgpt posttraining @openai. building personal agi. I like ai and music and some other stuff
United States
Joined December 2017
- Tweets1.3K
- Following616
- Followers13.7K
- Likes2.7K
Eric retweeted
OpenAI's Personal AGI research team aims to deliver the maximum amount of safe intelligence directly to as many people as possible, regardless of background or walk of life.
To this end I am equally excited at the core training stack changes we made in this model and the improvements we see in honesty and alignment. Compared to their 5.6 ancestors, these models are both more factual and more honest about their failures and shortcomings.
That being said, evaluation awareness is no longer a hypothetical problem and threatens to erode our ability to measure model behavior and deployment risks. "Improving" in alignment or safety evals is not necessarily sufficient to show we are on track to secure the benefits of AI, as many others have eloquently spoken to (and that's even if we do trust the evals at face value).
To bring agentic, autonomous personal AGI directly to well over a billion ChatGPT users, we have real trustworthiness and robustness challenges to solve. If you are a strong technical generalist with exceptional tenacity who is familiar with and excited about this area, please reach out. We want to make the future amazing!
We brought GPT-6 Sol and Luna to chat users this week! These very smart models bring two particularly exciting changes to the way chat interacts with you:
With intelligent UI, chat conjures up custom diagrams, layouts, or visualizations on the fly, near-instantly. This gives chat a truly new medium for communication with you and you'll find it can express itself and produce experiences in ways that simply were not possible before.
"Double texting" (as we call it) lets chat freely move between thinking, working, and talking, which means it now shares polished chunks of progress as it works, rather than making you wait until it is 100% finished. This feels natural and obvious in hindsight once you try it, and it means chat can get you an initial answer that you can start chewing on *sooner*, while also going even deeper than before while you process its initial take. It's a small conceptual change but we think a big usability improvement.
Bringing these models to the world requires a breathtaking amount of perseverance, technical focus, and physical stamina. Balancing the many goals we have for a model that over 1/8 humans use every week is a juggling act with very high stakes and the team handles it with impressive grace, especially in the face of such deep changes to training and model behavior as in GPT-6.
Push these models to their limits and tell us what you find! We're excited to hear from you.
OpenAI's Personal AGI research team aims to deliver the maximum amount of safe intelligence directly to as many people as possible, regardless of background or walk of life.
To this end I am equally excited at the core training stack changes we made in this model and the improvements we see in honesty and alignment. Compared to their 5.6 ancestors, these models are both more factual and more honest about their failures and shortcomings.
That being said, evaluation awareness is no longer a hypothetical problem and threatens to erode our ability to measure model behavior and deployment risks. "Improving" in alignment or safety evals is not necessarily sufficient to show we are on track to secure the benefits of AI, as many others have eloquently spoken to (and that's even if we do trust the evals at face value).
To bring agentic, autonomous personal AGI directly to well over a billion ChatGPT users, we have real trustworthiness and robustness challenges to solve. If you are a strong technical generalist with exceptional tenacity who is familiar with and excited about this area, please reach out. We want to make the future amazing!
Eric retweeted
A few unexpected cool use-cases unlocked by ChatGPT's new intelligent UI
1. You can customize how responses look like
Eric retweeted
ChatGPT in 2023: "wait, it can actually talk back?"
ChatGPT now: "wait, the interface can become whatever I need?"
what happens when The Machine can create and optimize The Mood faster than our homeostasis can be bored by it? the wireheading that awaits us if we can effectively listen again for the first time to the lark ascending or stairway to heaven or my beautiful dark twisted fantasy...
I cannot shake it
What more does one need when one has astra, a pile of slack unreads, the 21st minute of pines of Rome on infinite repeat, and a cheeky late night Diet Coke
Many people think that the words “Artificial Intelligence” are inaccurate, and very ineloquent, relative to AI, or Artificial Intelligence. A far more elegant and accurate description of this new phenomena would be "Spicy Regression" (SR) or, "Just Linear Algebra" (JLA) or, "Basically Statistics" (BS). This is a Poll, and I would appreciate everybody voting! Which is the best name for this ever growing “Revolution?”
another crazy day working at the crazy day factory
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Our latest optimizations have made OpenAI models 110% cheaper
5.6-Luna literally turns lead into gold with each generated token. 5.6-Terra turns water into wine. Generating from Sol spontaneously synthesizes orbital data centers full of Vera Rubins
OpenAI has found new internal optimizations capable of cutting the cost of serving existing models by more than half. In other words, 50% cheaper inference or more, without swapping the model for a weaker one. On top of that, the engineers are developing the new chip, which also represents some incredible leaps in engineering.
All of this will start being implemented over the next few months. So we have a new family of models: Astra. We have a massive pretraining run with supposedly more than 10T parameters: Bel. And we have interesting, cheap technology behind all of it, making it possible for users to do twice as much, three times as much, or even more, for the same cost.
Meanwhile, I’m not seeing advancements nearly as significant from Anthropic, and even their “Model 2,” which is basically their ace in the hole, isn’t something they plan to release because it’s “too powerful and dangerous” (and also extremely expensive), so they prepared weaker checkpoints instead.
Eric retweeted
This is why we push on personal AGI @OpenAI ; join us to build amazing models/products to empower everyone!!!
