Seth Gammon (@SethGammon) header
Seth Gammon (@SethGammon) profile picture

Advisor at @AnthropicAI Exploring the intersections of thought, action, and environment. Built Citadel, Harness for Claude Code + Codex

Joined September 2011
Photo by Seth Gammon (@SethGammon)Photo by Seth Gammon (@SethGammon)Photo by Seth Gammon (@SethGammon)Photo by Seth Gammon (@SethGammon)Photo by Seth Gammon (@SethGammon)Photo by Seth Gammon (@SethGammon)Photo by Seth Gammon (@SethGammon)Photo by Seth Gammon (@SethGammon)Photo by Seth Gammon (@SethGammon)
Seth Gammon retweeted
WHO WANTS MEDICARE FOR ALL?
2,392
3,286
359
33,097
619,445
Seth Gammon retweeted
We introduce 3D World Simulator, an interactive world model that runs at 50+ FPS on an RTX 5090. Our model predicts in a unified 3D representation space, capable of generating: • Novel view appearance • Multi-view consistent RGB-D • 3D geometry • Spatially aligned tactile signals Play with the world model now at jwang771.github.io/3dworldsi…!
57
163
43
1,753
164,870
Seth Gammon retweeted
Day 25/ We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today. We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage. And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
7,404
2,704
3,649
39,186
7,499,817
Seth Gammon retweeted
“Made with AI” triggers this instinct to devalue the output as fake. Like valuing an original painting versus a copy. The moment you're told it's AI, the value of what you're looking at changes, even if the work itself hasn't. And it's a visceral reaction. Suddenly something impressive becomes slop, something creative becomes lazy, something meaningful becomes artificial. There are plenty of legitimate discussions to be had about AI. Environmental concerns, economics, intellectual property, government, and everything in between. But I'm interested in this particular reaction, the immediate devaluation of something simply because AI was involved in making it. Because AI isn't copying in that same sense. Think of pencil, crayon, charcoal, or photography. Each has its own characteristics, its own way of producing something. AI is similar, both literally and figuratively. There's a style to how AI does things, shaped by how it's trained, configured, and instructed. When I make a story with AI, I'm not simply making the story I would've written myself. AI will make countless decisions I wouldn't, miss things I would've included, and introduce things I might never have considered. There's crossover, but the result is my idea interpreted through AI. And this applies to far more than writing or images. It applies to code, video, research, entire workflows, and everything in between. Take coding, for example. There are things people will never do entirely by themselves again because AI has made them accessible at a level of quality and convenience that changes the equation. You can design an entire application, make the decisions, shape the experience, work through the problems, and be deeply involved in every part of it while AI writes the code. But you'll never know exactly what that application would've become if you'd had to build every part of it yourself. The decisions you'd have made along the way, the problems you'd have encountered, the solutions you'd have discovered. That path doesn't exist. And the same is true in reverse. AI will make decisions you wouldn't have made, solve problems differently, and produce outcomes you might never have reached on your own. That's what I mean when I say AI is a style. Not just the appearance of something, but the way something comes into existence. Unlike a pencil left on a page, AI can continue working from an initial instruction, making decisions, using tools, and carrying out entire processes. You can configure how it operates, how much freedom it has, what it prioritizes, and how closely it follows your intentions. What you're doing is discovering and aligning a configuration that produces a particular outcome. Sometimes that means directing nearly every decision. Sometimes it means letting the system explore possibilities you wouldn't have considered. And I think philosophically, that's where some of this discomfort comes from. When someone says “Made with AI,” people imagine the human path that wasn't taken. The work you didn't personally perform, the skills you didn't have to exercise, the decisions you didn't have to make. And somehow that absence becomes evidence that the result is fake. But it isn't fake. It's also not the same thing as having done it entirely yourself, even if you directed every decision. It's something real, made through a different process, from a different vantage point. It's not inherently more, and it's not inherently less. It's made with AI.
2
1
5
125
Seth Gammon retweeted
A Showcase of this idea I'm messing around with, where a game session in Rocket League is saved, the theater file decoded and rebuilt to storyboard and stage clips edited for video. There's so much content generated by games that could have personal value, share with your friends, post as content. With a system like this you could storyboard a rivalry that took place throughout a match such as this one between myself and a user named Arthur.
3
2
3
154
Seth Gammon retweeted
Nothing regulates your nervous system like financial stability
101
1,727
55
13,535
116,477
Seth Gammon retweeted
i feel like Tibo signed up for the most stressful gauntlet I've ever seen in my life
24
2
159
5,288
Seth Gammon retweeted
You have to figure out how to remain optimistic. It's a difference of knowing that things can change, and wasting your time holding onto the safety that it won't. Be optimistic.
1
59
Seth Gammon retweeted
This is the worst time in history to have a "10-year plan"
231
452
91
4,902
113,161
Seth Gammon retweeted
brother, u need to hear this. YOU ARE THE MOAT
54
48
12
528
18,978
Seth Gammon retweeted
we should be more ambitious
136
379
67
11,728
1,434,313
Seth Gammon retweeted
Over the next 28 days, each day we’ll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset. Let the improvements begin.
All right, we’re locking in. Only things being worked on are simplifications, more efficiency for more usage, groundbreaking features or new models. Sometimes you have to invest ahead of the curve, but feedback is clear that you all want things to get simpler. On it.
4,481
1,492
2,378
26,295
8,301,697
Seth Gammon retweeted
If I ran a game studio, I wouldn't send out DMCAs. I'd be shipping SDKs. AI will decompile all the games. The only choice left is whether it happens with you or around you. Open source the code for anything older than a few years. Worried about piracy? id Software did this with Doom in 1997. It still runs on everything and people still buy it. Let the fans port your iconic gaming classics using assets from copies they purchased. You keep selling the game without paying for a remaster. Give modders a storefront and revenue share. Hire the best ones. Sega did that for Sonic Mania and got one of the best reviewed Sonic games of all time. Valve let Black Mesa sell on Steam. Fans get remasters sooner. Studios get to focus on shipping what's next, and IPs keep a community that's willing to pay for it. Fight your fans or fund them. Your move.
Every game ever put on a cartridge, disc, or download will get decompiled, modded, and ported with AI. If one dev and four Claude agents can decompile most of CoD MW2 (2009) in under a month, and Halo Combat Evolved (2001) is already running on browsers, phones, Macs, and Nintendo 3DS days after its decomp, AAA teams should be a bit scared. Fans won’t need to wait for remasters anymore. Studios can and will send DMCAs, but they won't outpace the frontier models being shipped. Open source gaming is here. Ready Player One's closer than we think.
39
63
8
598
17,695
Seth Gammon retweeted
Warn your family about this.
Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).
Readers added context they thought people might want to know
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard. cellcog.ai/blog/tavus-gri… research.nvidia.com/labs/amri/proj… tech-ish.com/2026/10/02/tav…
31
310
4
4,687
512,865
Seth Gammon retweeted
Being a dev today using frontier AI feels like I'm playing Watson to Sherlock Holmes. I just stand back and go "my word holmes how did you deduce that" and occasionally stop him from getting shot
31
665
38
11,064
225,002
Seth Gammon retweeted
Opus 5.5 had the best run with FishSlop that I've ever seen. The animation quality is incredible. The controls feel so much better than any other version I've generated. There's even a complete gameplay loop with a functional in-game economy and unlocks.
GPT-6 Astra is world class at Blender and 3 dimensional reasoning. This was a 1-shot game it created, all running in browser.
72
19
10
1,518
110,107
We are loading a BANKED RESET into all accounts of our Plus, Pro and Business users!!
Introducing GPT-6 Sol and Luna, bringing the advances behind Astra to faster, more affordable models. ✨ 💻 Stronger coding and computer use 🎯 Improved factuality and alignment 💬 Clearer answers with less jargon > On AutomationBench, Sol at xhigh effort outperforms Claude Opus 5 at max effort at just 9% of its cost per task. > On OSWorld 2.0 offline, Luna at max effort exceeds GPT-5.6 Sol at medium effort at one-tenth the cost. > Sol makes about half as many mistakes as GPT-5.6 Sol on our internal factuality evaluation of conversations where users previously flagged errors. > Improved prompt caching helps agents reuse more context, with 90% discounts on cached input-token reads. Developers can change reasoning effort and tool availability without breaking cache. Rolling out today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, and available in the OpenAI API. Free and Go users can access Luna in the desktop app. These models are not yet available in Chat. API pricing per 1M tokens: Sol: $2 input / $10 output Luna: $0.10 input / $0.50 output
110
23
12
768
53,033
GPT-6 Sol & Luna are ~50% cheaper than 5.6
Introducing GPT-6 Sol and Luna, bringing the advances behind Astra to faster, more affordable models. ✨ 💻 Stronger coding and computer use 🎯 Improved factuality and alignment 💬 Clearer answers with less jargon > On AutomationBench, Sol at xhigh effort outperforms Claude Opus 5 at max effort at just 9% of its cost per task. > On OSWorld 2.0 offline, Luna at max effort exceeds GPT-5.6 Sol at medium effort at one-tenth the cost. > Sol makes about half as many mistakes as GPT-5.6 Sol on our internal factuality evaluation of conversations where users previously flagged errors. > Improved prompt caching helps agents reuse more context, with 90% discounts on cached input-token reads. Developers can change reasoning effort and tool availability without breaking cache. Rolling out today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, and available in the OpenAI API. Free and Go users can access Luna in the desktop app. These models are not yet available in Chat. API pricing per 1M tokens: Sol: $2 input / $10 output Luna: $0.10 input / $0.50 output
36
13
4
366
22,822
Seth Gammon retweeted
If you blink you might miss it Jev + Computer use = 100x faster than any LLM @typesafeai
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution
39
59
9
1,277
127,652
Seth Gammon retweeted
holy cow guys luna maxxing with astra is actually great what the heck
89
13
10
1,636
186,873