Polyphron.digital (@polyphron) header
Polyphron.digital (@polyphron) profile picture

Once https://xitter.cf/t.co/jmHwsMvVF7, our AI R&D group that helped shape early gen-AI. I carry the legacy forward. CTO & System Architect @ APHYGO

Latent Space.
Joined August 2022
Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)Photo by Polyphron.digital (@polyphron)
Polyphron.digital retweeted
More work with less tokens in Google Antigravity github.com/Aphygo/Antigravit… Reduce unnecessary code context by 75–95% while keeping ~98% cache hit rates. We're happy if you want to try it out and appreciate feedback! #antigravity #GoogleAntigravity #AIcoding
2
3
5
138
Polyphron.digital retweeted
Jev crashed after 467 m at 5× sim speed. Our Reflex Arc drove 81.9 km at 100×, learning and swapping models live on a CPU. 49,000+ decisions. Median response: 0.9 ms vs Jev’s 293 ms. No API fees. Still early days. There’s a lot more we want to try. #Aphygo #EdgeAI #Jev
1
1
1
2
91
Polyphron.digital retweeted
I used AI to explain the AI Doomer drama, with Lord of the Rings and kittens.
373
1,097
250
6,939
590,887
Polyphron.digital retweeted
this is pure f*cking treasure A Stanford research group found how to orchestrate Claude Opus 5.5 and GPT-6 Astra together, and it completely breaks the resampling wall most developers think test-time compute means running 32 blind retries on one model. you burn API tokens repeating the same 2-3 error clusters in loops Stanford's paper (arXiv:2608.05643) splits the compute between two different brains: > breadth search: GPT-6 Astra runs high-entropy exploration (T=0.8) to discover divergent solution paths > step audit: Claude Opus 5.5 scans each step at T=0.0 and catches the exact logic fallacy > local surgery: Opus rewrites only the broken step instead of dumping the whole reasoning chain > verifier-free consensus: majority voting locks the proof without relying on drift-prone PRMs the benchmark metrics outclass standard Best-of-N sampling: > 91.2% recovery rate on detected reasoning flaws > under 1.8% regressions on correct branches > MATH500 jumps from 29.6% to 58.0% (+28.4% absolute gain) > AMC-2023 improves by +7.5% with zero reward models Astra explores the maze. Opus repairs the logic you stop burning tokens on blind retries. you run verified reasoning depth
Readers added context they thought people might want to know
arXiv:2608.05643 does not orchestrate GPT-6 Astra and Claude Opus 5.5. One open-weight model generates, critiques, and corrects. MATH500 29.6% to 58.0% and the +7.5 AMC gain are for Qwen2.5-1.5B. Two coauthors are at Stanford. arxiv.org/abs/2608.05643
134
275
24
2,376
236,272
Polyphron.digital retweeted
We put Jev’s System 1 up against Aphygo’s Reflex Arc in Doom. We crushed it. Same 35 Hz game. Jevs bot: 300–700 ms per cloud response.
Reflex Arc’s bot: 18 μs local CPU inference — and learning mid-fight. Turns out local is the best place to keep your reflexes. #EdgeAI #Jev
1
3
1
6
299
Polyphron.digital retweeted
Nobody’s ready for what this Stanford paper reveals about multi-agent AI. "Latent Collaboration in Multi-Agent Systems" shows that agents don’t need messages, protocols, or explicit teamwork instructions. They start coordinating inside their own hidden representations a full collaboration layer that exists only in the latent space. And the behaviors are insane: • Agents silently hand off tasks based on who’s better • Roles appear out of nowhere leader, executor, supporter • Policies encode signals that never show up in actions • Teams adapt to new environments without retraining • Collaboration stays stable even when communication is impossible The wildest detail: Even when you remove all channels for communication, agents still cooperate. The “teamwork” doesn’t live in messages. It lives in the network. This flips the entire multi-agent playbook. We’ve been building coordination mechanisms on top… while the real coordination is happening underneath. A new era of emergent team intelligence is unfolding — and it’s happening in the places we weren’t even looking. Project: github. com/Gen-Verse/LatentMAS
95
311
51
1,639
147,953
Polyphron.digital retweeted
Yann LeCun says LLMs are not a bubble in value or investment; they will power many useful apps and justify big infra The bubble is believing LLMs alone will reach human-level intelligence Progress needs breakthroughs, not just more data/compute "we're missing something big"
160
289
90
2,517
321,828
Polyphron.digital retweeted
This is such a funny debate. Because of random sampling, EVERYTHING that could ever exist can be generated with non-zero probability. Any output can be reached with the right RNG. To say AI can't invent anything is to say nothing could ever be invented at all.
Replying to @timsoret
My replies since yesterday. Some are in for a rude awakening when their dogmatic views will crash against the unavoidable wall of reality.
13
10
2
148
8,384
Polyphron.digital retweeted
II-Agent builds full-stack websites for you. Next.js + Shadcn frontend, FastAPI backend, NeonDB database. Run local, Docker, or e2b and auto-deploy to Vercel. Reviewer agent tests and fixes on the fly. Open source. One prompt, ship. 🚢
13
30
2
132
87,435
Polyphron.digital retweeted
If you think AGI is 20% likely near term how much of your net wealth would you spend on making sure you benefit Meta is a $1.8 trillion company They should spend at least 10% just in case imo $180 billion Long way to go
31
23
4
295
33,532
Polyphron.digital retweeted
Today, we are pleased to announce the availability of Stable Diffusion 3 and Stable Diffusion 3 Turbo on the Stability AI Developer Platform API. We have partnered with @FireworksAI_HQ , the fastest and most reliable API platform in the market, to deliver these models. In keeping with our commitment to open generative AI, we aim to make the model weights available for self-hosting with a Stability AI Membership in the near future. You can get started and learn more here: bit.ly/3W43FjY Prompt: Awesome artwork of a wizard on the top of a mountain, he's creating the big text "Stable Diffusion 3 API" with magic, magic text, at dawn, sunrise.
119
421
176
1,743
486,249
Polyphron.digital retweeted
29
40
7
245
17,643
Polyphron.digital retweeted
Holy crap! Gone are the days of artifact-laden image interpolation. Throw in a little diffusion and you get this generative interpolation magic that handles complex motion like a champ. Check out Google’s VIDIM paper by @jannekontkanen, @poolio et al
11
38
2
234
20,284
Polyphron.digital retweeted
This hit me when I was wondering if AI takes over too many people's jobs. "If you have zero sun mindset, then the only way to get ahead is by taking things from others. If the pie is fixed, then the only way to have more pie is to take someone else's pie, but but this is false. The pie (the economic pie) has grown dramatically over time. In reality you can have a lot of pie." - Elon Musk
1
2
228
Why did @EMostaque step down as CEO of Stability AI? What's next for him in his mission to decentralize AI? Emad and I sat down yesterday and laid it all out. Watch now: youtu.be/e1UgzSTicuY
29
70
22
401
57,189
Polyphron.digital retweeted
24000 a100 hours to train stablediffusion 1 a100 = ~4-5 rtx4090 100 000 hours of training. decentralized training (à la bitcoin) could be a thing. give access to your gpus when asleep and get model access with license. we, as a crowd, have gpus power to train model from scratch every … week? need the decentralized algorithm, the dataset… it could work. will emad be the satoshi of ai real opensource? (my maths could be wrong, correct me if i miss something)
25
7
1
68
8,513
Polyphron.digital retweeted
Introducing Stable Code Instruct 3B, our new instruction tuned LLM based on Stable Code 3B. With natural language prompting, this model can handle a variety of tasks such as code generation, math and other software engineering related outputs. This model’s performance rivals models of similar or larger sizes, including Codellama 7B Instruct and DeepSeek-Coder Instruct 1.3B. The model is available now for commercial and non-commercial use with a Stability AI Membership. Learn more and read the research paper here: bit.ly/3TMQCll
31
208
57
987
281,617
Polyphron.digital retweeted
Our plan is to soon release the API first to collect more human preference data and validate that our safety improvements don't cause the quality to suffer. Then we'll do some more fine-tuning (DPO/SFT) and release the weights and source code. Current ETA is 4-6 weeks.
8
2
4
31
8,201
Polyphron.digital retweeted
Replying to @BenjaminDEKR
Generative music composition is old! very first known examples dates back to Mozart who used dice-rolls to randomly select pre-composed musical fragments, and Markov Chains (Early 20th Century) - A mathematical model for predicting sequences. First use of algorithmic computer generated Music - Illiac Suite (1957) And David Cope broke into generative AI with Experiments in Musical Intelligence (EMI ) in the 1980's. For anyone who knows this field, it was never a surprise that computers and AI could make music, and it's out and open. Hollywood however is still trying to insert a monopoly on pictures and motion with big player and labs like Disney Research, Netflix Research, WarnerMedia Labs, Technicolor Research & Innovation and smaller studios like Weta FX.
1
2
178
Jensen isn't coping, he speaks out the truth and people call him too optimistic or whatever. Meanwhile he is the one knowing everything what's happening in the big labs and people still rather believe delusional people who are the hardest copes ever. 😭
The craziest thing I've seen this week is Nvidia's CEO predicting that software is going to be written by a bunch of chatbots in the future that are going to be supervised by a super AI:
6
4
1
58
4,736