Reg Saddler (@zaibatsu) header
Reg Saddler (@zaibatsu) profile picture

I build AI systems that have to show their work before they act. My tools publish their failures. Independent AI Safety Researcher & Founder, Difference Theory.

Minneapolis / TwinCities / MN
Joined August 2007
Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)Photo by Reg Saddler (@zaibatsu)
The FBI is investigating an apparent breach of its national crime database, and the USDA system that pays farmers was hit the same week. One agency has spoken on record. What were the police departments that use NCIC told, and when?
3
7
831
Reg Saddler retweeted
The luxury OpenAI used to have was that they could take another month and most of us would happily wait. Now that’s a month somebody spends using Claude (now even Gemini) for actual work. By the time OpenAI ships, that person might be perfectly happy where they are.
I want to know what the fuck is happening with openai’s leadership. You have some of the best researchers in the world, extraordinary compute resources, and enough money to pursue practically any research direction. Yet somehow the company is dealing with disappointing releases, growing internal pressure, researchers leaving, and an increasingly complicated safety process that’s reportedly delaying future models. Meanwhile google and anthropic are moving at full speed. How do you let an organization with that much talent end up here?
4
15
1,647
Reg Saddler retweeted
"AI just solved the hardest problem in math but I'm sure my mid-level accounting job is safe."
92
396
38
9,988
231,525
Reg Saddler retweeted
Everyone is talking about Gemini 4 Neon limits on the $100 Google AI Ultra plan and sleeping on the GPU receipt. Paying builders just ran the full $100/month stack: → 1,000 monthly Colab credits via CLI, roughly 55 H100 or 100 RTX6000 Pro hours → $40/month in Google Cloud credits → YouTube Premium Family plus 20TB storage Neither $100 Claude Max nor $200 ChatGPT Pro bundles raw H100 compute for your agents.
34
12
3
347
23,413
Gemini 4 Neon coded a ski simulator based on real Mont Blanc terrain. Built in @Antigravity with custom shaders and zero pre-made 3D models, it renders carved snow and avalanches live. Try it now → goo.gle/496Vk4f
250
208
99
3,747
201,495
Reg Saddler retweeted
Feedback for the @antigravity team if they want it Neon is truly awesome and a joy to use, and the harness keeps getting better so I'm closer and closer to using it as my main driver However, my number one complaint is that agy still goes into a dark hole for a long period of time, so I have no idea what is happening until things are over I think simply borrowing ideas from the Claude Code UX on communication what is happening, so the user can jump in if things are going in the wrong direction, would be a huge win
60
10
1
625
34,668
Reg Saddler retweeted
New NVIDIA paper finds that once an LLM has reasoned through a problem, the solution can run as CPU code for rephrased and re-numbered versions of it. that answers similar questions 16x to 291x cheaper. LLMs re-solve every problem from scratch, even when only wording or numbers change. Agents sending the same instructions with new data pay for that repeated reasoning each time. Their method has Claude Opus 4.5 turn a reasoning trace into small Python functions, with changeable inputs as parameters. At run time they execute on a CPU, calling a fine-tuned 1B model only for small text jobs like extracting values. The cache matched or beat the source model's accuracy on 7 of 8 math and science benchmarks, scoring 93.3% versus 83.3% on AIME '25. It handled at least 90% of rephrased or re-numbered variants. If your agent runs the same prompt on new inputs all day, cache the reasoning as code and use the LLM only on misses.
65
160
34
1,359
52,224
Reg Saddler retweeted
X quietly changed its algorithm 3 times this month... And the ranking code is public. I read it so you don't have to. Here's exactly what X rewards now, and how to make the most of it 🧵👇
37
13
9
134
20,228
Reg Saddler retweeted
For those of you trying out @antigravity for Neon, here's a quick tip: 1. Go to antigravity.google.com 2. Tap see more and then the share button 3. Tap Add to Home Screen Then you can talk to all your Antigravity threads via remote control on mobile. Even though Antigravity doesn't have a mobile app (yet), I find its remote control actually better than Claude's because all the threads are available by default.
Gemini 4 Neon is getting rave reviews, so I made a new video comparing it to Opus 5.5, my favorite model for everyday work. Some takeaways: 1. Opus is still better at 3D and design from my testing, but Neon is on par or better at writing and personality, which is a huge accomplishment. 2. @Google Gemini 4 has 3 models, but in reality I think I'll use Neon for 99% of tasks. It's much faster than Carbon and its results are just as good. 3. The $100/mo Gemini Ultra subscription is a steal. You get Gemini 4, YouTube Premium Family (my favorite subscription ever) for ad-free YouTube and Music, 20 TB of storage for all your photos and files, and more. 📌 Watch me test Neon vs Opus across 5 use cases here: youtu.be/iYIWuI5WfUo
35
16
9
258
48,111
Reg Saddler retweeted
How did I miss that I got an official Google reply omg
Replying to @theo
Glad you're liking Gemini 4 Neon :) Thanks for the review Theo, more good things on the way
98
70
6
8,085
190,796
Reg Saddler retweeted
1.1 billion output tokens for one claimed math proof. SemiAnalysis broke down both artifacts behind Anthropic researcher Levent Alpöge's Riemann Hypothesis claim at Fable 5.2's $40/M output rate: → Unreleased 250K+ page book: ~487M output tokens → Released 17.4M-line Lean proof: ~610M output tokens → Final output tokens alone: $44K → All-in with AgentX input/cache mix: $125K Factor in ~89% thinking and retry overhead from Anthropic's August math disclosure, and the bill hits $1.17M.
🚨🚨🚨ANOTHER MILLENNIUM PRIZE PROBLEM HAS BEEN CLAIMED🚨🚨🚨 Mathematician @__alpoge__ of @AnthropicAI claims that he used Claude to discover a proof of the Riemann Hypothesis, arguably the most significant open problem in mathematics. Experts are still verifying the released Lean proof, but it seems very likely correct. We estimate the output tokens alone would cost $44k at Fable 5.2 API prices, with plausible estimates for the total cost running in the millions. (1/4)🧵
16
2
1
37
1,874
Reg Saddler retweeted
After YEARS of using Twitter/X randomly and without purpose, I became intentional once again about a month ago. Here's the proof for all to see. Posting original content, engaging with others and being consistent makes all the difference. My first OCR should arrive today. I'm not expecting much. But I can say that it's good to be connecting again. And that's what it's really all about.
18
2
1
53
5,173
Reg Saddler retweeted
🎯🎯🎯
Claude didn’t solve video editing Claude can just make edits that look good If that’s what actually mattered then every editor on earth would be rich and famous I’ve hired 40+ editors and technical ability has never been the bottleneck The bottleneck is the high level creative decision making aka taste Ai editing is a game changer for solo creators with taste but will create 0 extra results for those without it Solve for the bottleneck if you want results
4
2
28
3,094
Reg Saddler retweeted
Claude didn’t solve video editing Claude can just make edits that look good If that’s what actually mattered then every editor on earth would be rich and famous I’ve hired 40+ editors and technical ability has never been the bottleneck The bottleneck is the high level creative decision making aka taste Ai editing is a game changer for solo creators with taste but will create 0 extra results for those without it Solve for the bottleneck if you want results
14
2
2
49
5,428
Reg Saddler retweeted
Reset everyone’s Gemini quota on Antigravity. The caching improvements have increased the amount you can get done in quota by 22%. Enjoy building with Neon over the weekend!
1,123
599
566
13,070
1,009,620
Reg Saddler retweeted
Volume on! Since I consider my AI tools coworkers, I thought I'd bring them together in a spoof of The Office. Here are the opening credits. These are the tools I use regularly. ChatGPT is buried in paperwork. Grok won't stop spinning in his chair. Claude's list of projects is never-ending. Perplexity gets into the weeds. Eleven Labs uses my cloned voice. Kling did the video you are looking at. Suno creates my music. Venice is sneaking out in a fake mustache. And Muse? Well we are just starting to play... I'm still the only one who battles with the copier. Send this to the coworker who uses ChatGPT for everything.
14
7
3
59
6,288
Reg Saddler retweeted
I’m quoting this post because I think it’s very important. Today changed a lot, and above all for education. The Riemann Hypothesis, the holy grail of mathematics: one mathematician, Claude and six weeks. If the proof holds, and the first checks look good, that’s all it took. After today it’s hard to argue that AI can’t do creative science at the highest level. The biggest challenge now is science itself, and how we train the next generation of scientists in math, computer science and STEM as a whole. You’ll still need deep expertise to understand what AI produces and to ask the right questions. But the scientist as the person who generates new knowledge will increasingly be replaced by AI systems that a few experts steer and check. And today’s models are the worst they’ll ever be. This immediately raises the question of what happens to university education. We won’t need scientists in the same numbers, for the same work or with the same training. Ever more capable models will produce scientific knowledge, and universities will lose part of their reason to exist. -We will generate knowledge on a scale that human scientists will struggle to evaluate. AI will take over much of both generating and falsifying it. -Universities, as places of higher learning and training for researchers, will give way to personal AI tutors. Knowledge becomes a commodity that people acquire as they see fit. -The result will be an explosion of scientific knowledge. The challenge then is to bring all of this into the real world. Robotics will play a key role there, and the next leap is already close. And yet most people still don’t understand what happened today or what to expect next year.
idk how to say this but Riemann being solved means the world after today is no longer the same world i grew up in. as a math student, the intellectual excitement of thinking about problems like this, the impact - that a lot of scientific workers attach their life’s meaning to - is gone. frontier scientific discovery now is just like tapping on a calculator. quoting a friend: after today, math class will become PE class. It’s only a way for kids to exercise their brains, a skill like a language, but no longer an essential subject to pursue and devote your life for cuz "you were making progress for humanity". lots of people’s life’s work is now just ashes in history. this marks a milestone where we can pretty much believe AI has been proven to be smarter than humans in any capacity, including frontier scientific discovery; and it will keep growing to be smarter and better at everything humans can do. physical ai is next. so many, many de facto social norms and foundational beliefs about a man's value are being redefined by AI progress. our societal value system - what can you do as a human that is unique, what’s your value to society and humanity as a whole and to others around you. what is fulfilling to achieve, when AI now can do everything better than us? we must soul-search for new meaning as a species. and before that consolidates, there will be riots and an uneasy phase. the progress brought by ai can be disruptive to our society's current structures, economy and labor market. people will need to reskill, and regimes need redesign on how value is distributed. i get the doom scenario now - humans may become 2nd class citizens and ai can enslave us if they find a reason to. how are we gonna compete with them in any labor or intellectual game?
71
88
8
1,014
62,560
And Vibo was born!
We’re rolling out a change to increase cache hit rate on Neon by around 4% on Antigravity by massive increasing the cache pool size. This will increase what you can get done in quota by at least 20%. Will issue a reset after it’s done since it burned quota faster!
1
1
932
Reg Saddler retweeted
53
80
21
1,726
300,025
Reg Saddler retweeted
Wait a second. If this plays out, who signs their names to the accounting at NVIDIA? The new interim CEO. What a mess.
BREAKING: The US Securities and Exchange Commission (SEC) is launching an investigation into "potential accounting issues" at Nvidia. Nvidia stock, $NVDA, extends losses to over -8% on the day and -45% since October 6th.
14
13
2
219
35,345