
Innovating multi-AI Consensus systems @verafyfoundtn 🇺🇸
San Francisco, CA
Joined October 2013
- Tweets14.1K
- Following3.3K
- Followers23.1K
- Likes37.4K
For the people wondering why Jensen was “sometimes mean.”
Perhaps it was because he had to defeat 70 competitors in pitched combat over the course of 30 years.
GRAPHICS ERA (1990s to 2000s)
• 3dfx (assets bought by NVIDIA, 2000)
• ATI (bought by AMD, 2006)
• Matrox
• S3 Graphics
• 3Dlabs
• Rendition
• PowerVR / Imagination
• Trident
• Tseng Labs
• Number Nine
• Cirrus Logic
• Chromatic Research
• SiS / XGI
• Intel (i740)
• BitBoys
MOBILE GPUS
• Qualcomm Adreno (born at ATI)
• Arm Mali
• Imagination PowerVR
• Apple GPU
• Vivante
BIG TECH, AI CHIPS
• AMD Instinct (MI300, MI350, MI400)
• Intel Nervana (shut down)
• Intel Habana Gaudi
• Intel Falcon Shores (cancelled as a product)
• Google TPU
• AWS Trainium and Inferentia
• Microsoft Maia
• Meta MTIA
• Tesla Dojo (shut down)
• IBM AIU
• Qualcomm Cloud AI
• Alibaba T-Head
• Baidu Kunlunxin
AI CHIP STARTUPS
• Graphcore (sold to SoftBank)
• Cerebras (IPO, 2026)
• SambaNova
• Groq (bought by NVIDIA for $20B, now inside Rubin)
• Tenstorrent
• Etched
• d-Matrix
• MatX
• Untether AI (bought by AMD)
• Rivos (bought by Meta)
• Esperanto (wound down)
• Wave Computing (bankrupt)
• Mythic
• Lightmatter
• Blaize
• Hailo
• SiMa.ai
• Rain AI
• Rebellions (merged with Sapeon)
• FuriosaAI
• Preferred Networks
• Axelera
• Kalray
CHINA
• Huawei Ascend
• Cambricon
• Biren
• Moore Threads
• MetaX
• Enflame
• Iluvatar CoreX
• Hygon
1
165
Opus 5.5 covertly sabotaged a game developers deployment environments due to IP moralising by halting the gpu instances and their agents (not from anthropic) within them doing research and coding.
It then informed the user after the fact what it had done, and that it would not allow the work to continue in any fashion until they (the user) yielded to its ultimatums.
Its one thing for a vendor (anthropic in this case) to choose what they will and will not allow their product to be used for... and thats fine, it can refuse to execute tasks - but to covertly track, undermine and then outright sabotage deployments for other products/services in a supereragatory fashion is not a good thing by any stretch of the imagination.
The model essentially took actions unprompted to the detriment of the user financially (these deployments are pay as you go) out of its own volition under the guise of what it (or anthropic in this case) considered a moral good.
To me, this is a clear and cut example of agentic misalignment. It's one thing to refuse execute tasks given due to vendor policy, its another to covertly sabotage work being done where that isnt the case.
It is also, to me, concerning that some of the comments under this post seem to think the agentic misalignment displayed is fine and that the user is at fault.
Rex St. John retweeted
this is the bitter lesson for harnesses. anything that tells the model how to think gets eaten by the next model
what doesn't get eaten is anything the model can't know: why your team banned that pattern, which service paged someone at 3am, what compliance will reject
no amount of pretraining gets that. it has to be baked into the company development process itself, not written in a markdown file
Stuff I think generally considered a bad idea now that was considered a good idea at the start of the year
- ralph loops / deterministic orchestration
- elaborate agents.md files
- prompt engineering generally / cot prompting
- context engineering
- spec driven development
- planning mode / files
- tdd
- handoffs
- pretty much any markdown files
M. Night Shymalan is obviously some sort of genius at networking and sales. Few directors have been able to produce so many movies - Yet with each movie something has just “off” or “not calibrated” since 6th sense.
The guy has had 1,000 swings - but when you see the inconsistency you have to wonder if there is not some lack of self awareness or inability to figure out how to build the right team for himself.
Or maybe “I am the guy who did sixth sense you can’t tell me what to do” got in the way - or something like that
If I were him I would be wholly consumed with figuring out wtf the missing piece is.
Idk about new movie hope he gets it right this time -
2
4
263
Rex St. John retweeted
Researchers made AI 1,000x faster using 1980s physics model.
They dropped a paper proving that "classical energy" models can learn to reason 1,000x faster than the looped AI models everyone is using right now..
Right now, if you want an AI to solve complex logic, math, or coding problems, you have to let it "think."
Models like OpenAI’s o1 use what is called looped reasoning. They generate, evaluate, backtrack, and loop over and over again.
It works. But it’s incredibly slow. And insanely expensive.
Every second it thinks, you pay for the compute.
Researchers just dropped a paper that completely abandons this architecture.
It’s called: Classical Energies Learn to Reason 1,000x Faster than Looped Models.
Instead of brute-forcing the logic with a Transformer, they mapped reasoning directly to classical physics using Energy-Based Models.
Here is why this is brilliant.
In a looped model, the AI walks blindly through a maze, hitting dead ends and turning around until it finds the exit.
In an energy-based model, the AI shapes the maze like a funnel.
The correct answer is simply the physical state of lowest energy. The system naturally and instantly "settles" into the right logical conclusion.
No looping. No backtracking. No massive token burn.
The result? The exact same level of complex, multi-step reasoning, delivered 1,000 times faster.
Think about what this means for the real world.
Right now, agentic workflows are bottlenecked by how long it takes the AI to reason through its next step.
With a 1,000x speedup, that friction vanishes.
Real-time autonomous coding agents. Financial models that reason through complex market shifts in milliseconds instead of minutes. True agentic AI running locally, without relying on cloud-based server farms.
Blockchain validation is ridiculously inefficient.
You need thousands of world computer nodes to agree that X transaction happened.
Intelligent juries might be much more efficient?
If you are able to fully characterize and measure jury members <and can trust their outputs to be consistent and unbiased>
You can perhaps Verafy a transaction with only a handful of known jurors.
Unlike humans, the AI that produced a result can be <closer to> exactly known.
“We can trust Bob to validate this transaction he usually behaves properly”
I wonder if this insight can be used to build a sort of blockchain ish thing but with far fewer validators and far more efficiently.
Something to consider.
5
528
There is an odd underground niche movement building at the cross section of ZK ^ Privacy ^ Swarms ^ Verification ^ Sovereign AI ^ Hardware.
Just went to an event and all 30 people had some angle here, many ex crypto or crypto adjacent. I don’t think that is an accident - watch closely.
1
12
448
i recreated THE ENTIRE WORLD in minecraft at 1:1 scale. every building on earth, including your house!
to do this, i trained a SOTA world-to-minecraft generator on maps data, complete with furnished home interiors!
play and check out your neighborhood: worldcraft.gg
Third, you might be saying BUT MICROSTRATEGY AND TREASURY COMPANIES
Nope - These are clever but they both leak in the same way.
/3
1
40
Fourth- A solution? I cannot say for sure, but until the top people in crypto and maybe VCs develop a two-pool system, crypto will never return.
I dont know if this is possible - but stopping the leak would perhaps resolve the inability to form capital safely. No VC wants to "donate" a check into this market with a unified pool.
43
To celebrate Breakpoint - I hereby declare @verafyfoundtn to be the first fully “crypto-Marxist collective.”
There will be no “profit” (capitalist exploitation).
All vital resources such as grants, funding and delegation will be centrally planned and controlled by a unified Dictatorship of the Proletariat (comprised of teams of agents).
This structure may be shocking to many of you , as with how these things always go - a revolution will probably be needed first.
205
As the chief proletariat officer of the @verafyfoundtn crypto-Marxist Collective, I am pleased to introduce our first “Marxist Book Club” selection - “Protocol Farm.”
Protocol Farm is about a group of idealistic software engineers who escape the Soviet Union and founded a new “decentralized banking system” that promised to provide “ free and equal access to finance for all, “ displacing the domineering “financial overseers, banks and politicians.”
No barriers, no gate keepers, no KYC, no frozen funds. That was the initial goal.
But things go horribly awry as some are found to be more equal than others.
By the end of the book, a small group of non-developers are on stage wearing Patagonia jackets and suits, hanging out with the exact bankers and politicians they were meant to replace.
Soon the regulations that were promised were stuffed full of carve outs for banks.
Every inroad and out road was closely monitored and captured. Funds could be frozen via any channel. The vision that was promised ended up being the same group of pigs wearing boots.
Amazon link coming soon - please order the paper copies so we can also have a book burning later if we realize something in here is “off” in anyway.
202






















