IM
Gpt 6 Astra is pre AGI and bel is AGI
Feel The AGI1,910 posts
FIELD NOTES · 2020—2026
Feel The
AGI
Not a feed. A time capsule of technological acceleration, captured while it was still happening.
1,910 bookmarked fragments · reverse chronological
About this archive
ABOUT THE COLLECTION
Living through the winding-up of the world.
AGI, robotics, medicine, China, labor, surveillance, drones, markets, and the odd little jokes people made when they felt the ground move.
LE
🚨 BREAKING: "gpt-6-astra" has been staged on the OpenAI API The OpenAI Responses API now returns a 404 Not Found for "gpt-6-astra", where garbage/actually non-existent slugs return 400s - a 404 is also returned for 5.6 Cyber, which we know exists. A similar quirk exists in the Bedrock API used to find Fable 5.1 a day prior to launch.
Per The Information, although the name "GPT-6" was considered for Astra, they decided against calling it that That narrows down the possibilities essentially just to "GPT-5.7 Astra", or just "GPT-Astra". And it suggests they're planning for Bel to be the 6-worthy jump
AR
New ARC Prize 2026 - ARC-AGI-2 High Score 73.33% by Rabbithole (new 1st place)
LO
Introducing Gemini 3.8 Flash, another jump in Gemini's agentic + coding capabilities, and our 3rd updated Flash model in only 6 weeks... This model has been a ton of fun to work with, excited to see what you all think!
WE
I am begging you to stop using AI to write your video scripts. Shortform content is becoming unwatchable and that's not even the crazy part
UN
Economic status has cratered for young less-educated men, per FT:
EL
We're organising a competition for AI-generated philosophy with $11,000 in prizes. Consider entering!
AS
Fable 5.1 is cracked. This was my first test with Fable 5.1. I asked it to create an F-35A Lightning II, and I honestly never expected it to be this good. Every single thing you see here was created in code.
JE
The funny thing about the 'takers and makers' paradigm is that rural conservative areas are overwhelmingly dependent on government handouts while liberal cities fund the country.
Our great national tragedy: "Today, most U.S. counties depend on a level of government transfer income that was once reserved only for the most distressed places. In 1970, not even 1 percent of counties derived a quarter or more of their total personal income from transfers. In 2000, just 10.4 percent did. But by 2022, 53 percent of counties were receiving a quarter or more of their income from transfers." - @InnovateEconomy
CH
Holy, Astra is build different: OpenAI’s Astra model reportedly uses “recurrent depth” to improve coding and computer-use performance, by sacrificing visibility into how AI reasons. Recurrent depth processes text through the same model layers multiple times, potentially allowing smaller models to perform like much larger ones while reducing memory and bandwidth costs. OpenAI has reportedly limited Astra’s use of the technique so researchers can still monitor its chain of thought. The company says Astra will launch with additional monitoring to rapidly detect and contain misbehavior. The concern extends beyond Astra: unrestricted latent reasoning could make future models more capable while removing one of the few windows researchers have into dangerous behavior. Tuesday night, OpenAI Chief Scientist Jakub Pachocki said in a post on X that although monitoring of models’ chains of thought was “fragile” and “unfortunately heading in a negative direction.
OpenAI’s Astra AI uses a new reasoning approach called “recurrent depth.” Though it can help model costs and performance, researchers are concerned bc it obscures a model’s thinking process, making it more difficult to monitor. w/ @amir @rocketalignment theinformation.com/articles/secre…
RH
remember how simple the dev world used to be that this was a big drama
Breaking: @dhh follows @theo. World peace ensues. But seriously @lexfridman consider this my vote for getting Theo on the show 👀
RI
PO
the corny oriental music it starts playing when it stances up 😭 i'm crying
Rusya'da bir teknoloji mağazasında robot, kendisini iten müşteriye saldırmaya çalıştı.
AI
Things are now happening *much faster* than AI 2027 AI 2027: In March 2027, a reckless AI company will invent neuralese (basically, we lose the ability to monitor AIs thoughts) REALITY: Summer of 2026 WHY THIS A "HOLY SHIT FUCK" MOMENT: OpenAI itself - and basically everyone in AI safety - warned against doing this exact same thing 10 months ago in a open letter and blog posts. The AI safety community is livid. Currently, models reason in plain English, so we can read their thoughts to see if they're scheming against us. This is how we lose that ability forever. It sets a very dangerous precedent.
OpenAI: At long last, we have created The Neuralese Model from the classic OpenAI blogpost from 10 months ago, Don't Create The Neuralese Model.
MA
More powerful but also more opaque, what could go wrong?
new: OpenAI & others quietly using loop transformers that don't show their 'thinking' when scaled up a leap forward on performance, but sparking concerns inside & outside OpenAI re: security as this takes off
CR
This data center is - Creating a police department - Lowering garbage bills - Funding community programs - AND ALSO giving out $10,000 checks When can I get a data center in my backyard?
A Pennsylvania data center is offering $10,000 checks to residents as part of its benefits package for the community
VX
Dawg, they got Pete Hegseth in the data breach. It's available for sale for $100
I really recommending reading this. In summary, a company which does ID verification for in-person interactions (hotels, car rentals, ID verification for alcohol or marijuana, etc) has some how exposed over 153,000,000 drivers licenses for people in the United States and Canada. It is a catastrophic data breach, probably one of the worse I've ever seen. If you're in the United States and have traveled, gotten a hotel, purchased marijuana or alcohol, there is a high probability you're in this. Unlike other breaches, this includes a photo of the person (from the license), making verification you've identified the person significantly easier. This poses a significant threat to celebrities (musicians, YouTubers, streamers, adult entertainers, actors, etc), politicians, lawyers, wealthy people (CEOs, investors, people of public interest), Law Enforcement Officers, etc Krebs himself, and several other security researchers, have already confirmed they're in the data leak. tl;dr gah damn dawg this company is going to be sued into oblivion krebsonsecurity.com/2026/09/fbi-pr…
DA
Claude can just log into your wifi for you and make it better
VX
I really recommending reading this. In summary, a company which does ID verification for in-person interactions (hotels, car rentals, ID verification for alcohol or marijuana, etc) has some how exposed over 153,000,000 drivers licenses for people in the United States and Canada. It is a catastrophic data breach, probably one of the worse I've ever seen. If you're in the United States and have traveled, gotten a hotel, purchased marijuana or alcohol, there is a high probability you're in this. Unlike other breaches, this includes a photo of the person (from the license), making verification you've identified the person significantly easier. This poses a significant threat to celebrities (musicians, YouTubers, streamers, adult entertainers, actors, etc), politicians, lawyers, wealthy people (CEOs, investors, people of public interest), Law Enforcement Officers, etc Krebs himself, and several other security researchers, have already confirmed they're in the data leak. tl;dr gah damn dawg this company is going to be sued into oblivion krebsonsecurity.com/2026/09/fbi-pr…
BE
A professor at Colorado University told his class this week: “If you plan to use AI in this course, tell me now. I’ll give you a B. You don’t have to attend, take a test, or do any work. If I catch you using it and you didn’t take the deal, you get an F.” The class is called Critical Thinking.
SA
Over the summer, we have been sprinting on safety priorities; it's more important than ever for capabilities and safeguards to advance together. We have more to do but have made a lot of progress. We are also going to be launching our next model soon. There is an obvious tension here: on one hand, Astra is very good and we are excited to see what people will build with it. We are proud of our work. On the other hand, we are clearly in a phase of development where we believe caution is warranted, and we are pacing our progress to ensure that we can meet the safety standards required by new capability levels. Astra has been done training for a while now and is a significant step forward in both capabilities and alignment. For the models after that, we have been slowing things as needed to ensure that we can do sufficient work on safety and alignment. AI is getting extremely capable; no one fully understands the consequences of this. Managing the transition to a world with abundant and powerful AI to optimize for safety and benefits to people should be one of the highest priorities in the world. It is our highest priority at OpenAI. We have been living with the tension between being excited and anxious about progress for some time, and it is still discordant for us. We know it is much more discordant for other people. And yet, we believe strongly that the world needs to understand where AI is going and how models perform in the real world. More importantly, we believe the world will need aligned AI to manage the future phases of this transition. An iterative loop where society and this technology evolve together is what will lead to the highest chance of getting this right. So we hope you enjoy our new model, and we hope the world continues to take what’s happening in AI extremely seriously.
NO
Climate models have been extremely accurate, even going back to the 1970s.
We need to reckon with the fact that we were all sold decades of catastrophizing over climate change and none of the predictions came true.
RY
still don't know what to make of all this i used to write software for myself like this all the time but it took way longer than a dictated text message to a grok bot and i kinda cared about the code
MI
AI video is 100% solved...
Introducing Higgsfield Genjutsu. Our strongest AI video transformation tool yet. Upload your video, add your character, and transfer the exact motion in a few clicks: acting, lip sync, camera movement, and VFX, with every detail held consistent. Try it for free on Higgsfield.
SE
Guys there’s genuinely a height requirement of 5’9” minimum to donate sperm. Holy blackpill
BE
Love this. We've merged 10 skill improvement PRs so far by automating self-improvement, more in review. Here's one case the self-improvement agent caught, where our agent was wasting tokens polling its orchestrator after a task was complete. These costs add up!
AR
Fable 5.1 from @AnthropicAI on ARC-AGI (Verified): - ARC-AGI-2: 90.0%, $3.12/task - ARC-AGI-1: 97.5%, $1.40/task Across the two benchmarks, Fable 5.1's average cost per task is about 32% lower than Fable 5's, driven by better token efficiency.
BE
I'm particularly excited by Atlas's ability to reconstruct scenes from a very small number of input image -- was able to create this flythrough of London's Natural History Museum by combining 3 input images I found from completely separate sources on Google Images
TE
Perhaps the clearest demonstration of the divergence in Anthropic and OpenAI priorities. I think OpenAI is correct.
Claude Fable 5.1 tops the Artificial Analysis Intelligence Index but costs 20% more per task than Fable 5 despite a 75% cache read price cut We supported @AnthropicAI with pre-release evaluation of Claude Fable 5.1. At max effort it scores 66 on the Artificial Analysis Intelligence Index, the highest score we have measured, ahead of Claude Opus 5 (max, 63), Claude Fable 5 (max, 62), GPT-5.6 Sol (max, 61) and Grok 4.6 (high, 61). We evaluated the model with Anthropic's ‘default’ server-side fallback, which routes safety-flagged requests to Claude Opus 4.8 or Claude Opus 5; fallback served ~4% of output tokens across the Intelligence Index. Key takeaways ➤ Frontier Intelligence with improvements across benchmarks: Fable 5.1 gains +4 points on the Intelligence Index over Fable 5. On HLE, Fable 5.1 scores 59.1%, ahead of the previous best of 55.5% from Claude Fable 5. It posts the narrowly highest scores we’ve seen on Terminal-Bench v2.1 (91.4%) and SciCode (62.0%), and on τ³-Banking it gains 9 points over Fable 5 ➤ 75% cache read price cut, but Fable 5.1 still costs more per task: Anthropic has cut the cache read price from $1 to $0.25 per 1M cached input tokens, with standard pricing unchanged at $10/$50 per 1M input/output tokens. Fable 5.1 (max) costs $3.76 per Intelligence Index task, 20% more than Fable 5 (max), because it uses ~1.7x the output tokens. The cache cut saves ~$1.40 per task, concentrated in the agentic evaluations where the majority of input tokens are cache reads. At xhigh effort Fable 5.1 scores 65 at $2.72 per task, $1.04 less than max, but still above Claude Opus 5 (max, 63) at $2.34 ➤ Claude Fable 5.1 holds the upper end of the Intelligence vs Output Tokens per Task Pareto frontier: every model variant scoring higher than GPT-5.6 Sol (medium) on the Intelligence Index is matched or beaten by a Fable 5.1 effort level on both intelligence and token usage ➤ Highest scores on agentic work tasks, but effectively tied with Opus 5: Fable 5.1 sets the highest scores we have measured on GDPval-AA v2 (1,853 Elo, +130 over Fable 5) and AA-Briefcase (1,694 Elo, +122 over Fable 5), our agentic knowledge work evaluations. Against Claude Opus 5 the GDPval-AA v2 lead is within the confidence interval and AA-Briefcase (1,685) is effectively tied, with Fable 5.1 ahead on analytical quality and rubric correctness, but behind on presentation Other model details: ➤ Context window: 1 million tokens, supporting image and text inputs as with Anthropic’s other recent launches ➤ Pricing: Fable 5.1 retains the $10/$50/$12.5 input, output, and cache write prices per million tokens from Fable 5, but cache hits have been reduced to $0.25 per million tokens, a 75% relative reduction from before that will materially reduce agentic workload costs
RI
Milo: "I didn't really understand the physical reality of how institutionalized the brutality is in America.” It sounds like this has given him a lot to think about. I hope that he reflects on it and decides to be a positive voice for change.