Crypto. AI Agents. TCGs. NHI. ✝️ Professional Trading Card Printing Slack + Discord Automation Pro Community Vibe Curator Streamer
Joined December 2023
- Tweets21.1K
- Following11.8K
- Followers16K
- Likes75K
Pinned Tweet
Very proud to announce you can now connect to @TCGPlayTest MCP server and order custom cards easier than ever before
Check it out here:
tcgplaytest.com/api#mcp
The TCGPlaytest MCP server is here. 35 cents per card!
It has never been easier to create and order your own custom cards
Check it out:
partner.tcgplaytest.com/docs…
I think @elonmusk should build the robotics factories underground because these androids will form the armies that keep technology progressing... surface factories will be too vulnerable. Musk will need these armies to protect the assets that deliver the payloads to space... once in space they are largely invulnerable. Data centers now... eventually lasers and other DEWs. A network of satellites that orbit Earth 5 times a day is a lot easier to aim at virtually any target. The little aircraft carrier that takes 5 days to reach a target is a joke. Eventually moon based operations facilities that can airdrop supplies anywhere on Earth to facilities protected by android armies that build each other 24/7. This is the next decade.
If you’re mad the left won… who cares. If the republicans won the left still won. We dont actually have a right wing
If theres a multiverse with constant splitting timelines then this is defacto incorrect as we would be in the version where this isnt the case
Nick retweeted
🚨ANTHROPIC IS USING CRYPTO-STYLE ZK PROOFS FOR AI INFERENCE!
Anthropic just open-sourced FlashProve, a system that generates zkSNARK proofs to show a specific model produced a given response without revealing the weights.
The proof system is the same kind used in Ethereum L2s such as Starknet and Mantle.
Collaborators include Alessandro Chiesa (StarkWare/EPFL), a co-inventor of Zerocash and co-founder of Zcash, plus people from Succinct, MIT, and EPFL.
This brings zero-knowledge tech that was battle-tested in crypto directly into verifiable AI outputs.
Nick retweeted
This is so insanely misleading.
No, it did not edit the raw weights. It used existing tooling to identify a problem area for the model, then it found the direction associated with it, and ablated that from the model.
The oversell here is downright intellectually dishonest.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Nick retweeted
Anthropic has a 15-person chip design team and has told investors they are using Claude to design chips faster than Nvidia or AMD
Earlier this year at GTC, Nvidia’s Chief Scienst Bill Dally told Jeff Dean how Nvidia uses AI to speed up the chip design process:
▫️trained an LLM on all proprietary internal Nvidia docs over past 30+ years (junior employees query it instead of interrupting senior designers)
▫️one AI tool ports Nvidia’s cell library to a new semiconductor process and does it in only one night (used to take 10 employees up to 8 months, or 80 total person-months)
▫️since the 1950s, there’s been a classic chip design problem of where to place look ahead stages in a chain (AI is coming up with solutions using “bizarre designs thay no human” would think of)
▫️agentic AI systems are doing a ton of exploration…testing parameters spaces…suggesting new approaches…running architecture experiments
▫️verification process is laborious but AI able to do it at fraction of time (compresses time from design to tape-out, with TSMC making chip)
Dally says it’s still a long way from end-to-end chip design but imagines a world where one master AI agents manages multiple sub-agents (similar to the current human-led process).
Anthropic’s 15-person chip design team is using Claude to develop its own AI chips, The Information reports.
The team has found that Claude speeds up the work compared with its estimates of traditional chip design workflows.
In doing so, Anthropic is following the path taken by OpenAI, which has designed its own chip, codenamed "Jalapeño."
Nick retweeted
> Railway gets caught hacking competitor's infra
> “sorry, my agents accidentally did it”
This is insane
Hey @Railway, is there a reason employee's of yours are trying to breach my infrastructure?
Your employee Jake Runzer, just attempted to tonight.
Nick retweeted
We introduced the Gold Floor last week and it's already proved itself to be a powerful mechanism.
Thoughts:
1. Basically we reshaped the $GLD in the treasury that has accumulated thus far into a massive robinhood:0xfce34d21a92069100e3226638140706cf03a266e buywall at support. Currently it could absorb around 20% of the entire floating supply, we're targeting 25%, sustainable and perpetually increasing.
2. Three days after we announced the gold floor, binance was hacked, we couldn't have manufactered a better stress test in a lab. It held strong then bounced while most onchain tokens bled and continue to bleed.
3. As the price of Gold increases, not only does the bid increase, but the floor price increases as well. So if Gold were to double from here, the $150k bid at $5.8m FDV becomes $300k at over $13m. That's if we add nothing and do nothing.
4. As swap fees collect and the treasury collects revenue, they get added to the top of the floor in bands, $20k bids per $100k mcap, so the floor increases from $5.8m -> $5.9m -> $6m and on... slowly but surely, defending higher lows.
5. The floor will be rebalanced periodically, and the bottom bands will be shifted up to the top, racheting up the floor, so long as the target 25% of floating supply is maintained.
6. Tomorrow is our mint and app launch. Mint funds will go towards bootstrapping gFUNDs and increasing the gold floor, we could potentially add another $100k in bids to the gold floor tomorrow just with mint proceeds.
7. Users will also begin staking their robinhood:0xfce34d21a92069100e3226638140706cf03a266e for the first time tomorrow, taking float off the market, and increasing the fractional backing ratio, allowing us to rachet up the floor further. I'm speculating but we could see 30%+ of float staked in the first 48 hours post launch.
8. Publicly announcing your floor has benefits and risks, one of the benefits is how clear it shows new users waiting for an entry to bid, or those looking to accumulate. Place your bids right above the floor, there's your entry. This creates a loop alongside the open market as new bidders further reinforce the support level.
9. You'll never hear me say the gold floor is unbreakable, I want that to be clear, I've been around long enough and seen enough in crypto never to claim anything like that. The goal is to give holders confidence, the treasury is bidding support.
AI gains showed up when tokens stopped being rationed
Makes sense to me. AI can easily demonstrate the capability to replace high paying jobs and services. With a single prompt… “produce an architecturally sound blueprint of this house that follows all the local codes and conventions” - Boom - it does it better, faster, and cheaper than a paid employee/human by multiple orders of magnitude.
Thats why people are so bullish on it.
So why didnt this immediately kill all jobs as fast as people like @DarioAmodei said it would?
It is because the workflow development and experimentation/refining cycles required to actually incorporate these capabilities into existing business systems and liability structures is the hard part.
Demonstration of the end product doesnt equal the entire job.
Integration of these capabilities into existing businesses to reduce costs and increase productivity is possible but it requires a lot of iterating… you need lots of human approval steps… you need layers of systems of checks and balances… what information should be surfaced by the AI and presented to the human and where/when should that happen. Building in acceptable levels of redundancy…. I could go on for a while.
This is the stuff that was skipped when these assumptions were made.
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list.
Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%.
Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything.
Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO.
h/t @OliverMolander
Would you move your team off frontier APIs if the bill dropped 90%?
I was thinking the same thing
I’m not at all shy about blaming DEI where appropriate, and it often is appropriate. But I see no DEI angle on this jet crash. The female pilot was not flying the plane. The captain was a 55 year old man. Trying to blame the girl for the crash is absurd and baseless. Yes she seemed frantic and terrified in the radio call but I would be to, and so would you. And even if she’d been cool as a cucumber, there isn’t the slightest reason to think it would have made any difference at all.
I’ll fully admit that I wouldn’t get in a plane if the copilot was a 20 year old girl. But that doesn’t change the fact that there is no evidence that she was personally at fault. And for the record, becoming a pilot at 20 is pretty impressive. She had her shit together a lot more than I did at that age. I’ll say that at least. I don’t think airlines or private charter companies should be making any specific effort at all to be “inclusive” of women or whatever, but it’s still admirable for a kid to have a goal like that and pursue it. I don’t see any reason to mock her, as some have done.
Some of the comments about that audio recording have been callous to an unhuman degree. None of you would have the stoicism in that situation that you demand from this girl. And again, there’s no reason to think stoicism would have magically kept the plane from falling out of the sky. Besides, she actually did seem pretty calm in that recording until the last few seconds, relative to the circumstance she was in. I don’t want my pilot to panic but I also don’t want them to be completely indifferent to the prospect of a fatal crash, as some commenters here apparently expect.
Why do almost all crypto currencies fail even if they were a good project?
there simply isnt enough value tied to them. Actually tying a value to the token itself is not easy, and most implementations are quite weak at best. Even some big ones with high market caps derive a lot of their value from the brand itself and not the actual token utility.
"Does this project need a token" is always a good question to ask to see if it has any sort of value beyond reputation and hype.
If a crypto currency is useful in some way to the project.. i ask "could this just use eth or sol" and "does the mechanism demand its own token" or "Does this project/company have enough utility to justify a token that gives it an independent valuation vehicle" basically some companies or projects are useful enough to where even if they could run without their own token... they sort of deserve a way to value it and a crypto with some utility can provide that vehicle.
Crypto is still at a stage where most of the value capture is still derived from trading fees for the vast majority of successful things. Hype, uniswap, the blockchains themselves, etc... The revenue is produced from transactions occurring. Whereas the real world value is generated from lots of stuff totally different like sale of goods and services.
AI has sort of given crypto another way to capture value... so im always keeping an eye out for innovative value capture/generation mechanics that leverage AI/Crypto.
Alot of projects that provide some sort of tech utility/service would probably actually have done better if they just didnt have a coin and just marketed the thing traditionally. So im always cautious because many people tried that... or just dont want to put the work in that that would require... so they take some half baked idea and attach a crypto to it and make like 30k off the fees they get from the initial hype spike. Some people had good intentions and it was just not viable... many know exactly what theyre doing though and just launch a token to suck as much value out of some vague concept of an idea that they can make sound decent... or they know people will buy based on their connections or reputation/following.
So yeah... always be a cynic and get good at judging viability.
Jfc now my inbox is full of people asking me how to quit their porn addiction and stop jerking off
- First... quit being a pussy. You are mentally stronger than you think... you should see yourself that way.
- Second... remove things that trigger that urge... clean up your social media feed... stop using certain apps etc
- Third... alot of it is boredom...
you arent actually doing anything productive that you are interested in... you need to start doing something... Ideally something that can earn you money at some point somehow.
Start learning how to use AI to research and develop shit. Just pick anything anything and go with it until it either becomes something or you lose interest and then move on to the next thing. If you curate the info you consume from social media/internet so that it makes you learn things and stimulates curiousity you will have no shortage of stuff you find interesting. Strive to progressively clean up your information consumption diet. Use AI and get good with it. Cut out unproductive habits like video games. I stopped playing games... i would only play them now if my life was set and i was rich. Literally they are a waste of fucking time.
Eventually, you will find something that you are naturally interested in and you will develop this sort of "pure drive" where youre like holy shit i just want to do this and its productive and i actually look forward to waking up because i want to keep doing it. You may get bored and exhaust that interest but because youve cleaned up your information consumption and learned how to use AI you can just jump to the next curiousity.
You eventually just dont have urges or interest in that degenerate shit at all anymore.
Its funny, i quit porn… masturbation… 1.5 years ago (very disgusting and energy sapping vices and everyone knows it) i have none of those impulsive degenerate behaviors anymore. And when i say i quit… i literally mean zero transgressions. I cured myself from this affliction through sheer will power (religious reasons for restraint is a crutch imo). In that same timeframe i became much more productive and co-founded a profitable startup (currently quarter mill annualized revenue)
Anyways these bad compulsions do much harm. This energy should be saved for your significant other, or converted into productivity. Stay hard on yourself… shame yourself, hold yourself to the highest of standards.
Quadrillion param models to match human neural circuitry 👀
Humans are ~1,000,000x more sample efficient than LLMs, and there are no good explanations.
We’ve found that a big part of the answer is that LLMs are *massively* underparameterized by many orders of magnitude. The brain has 100T synapses learning from ~200M tokens of language. Chinchilla prescribes a 1T-parameter model on 20T tokens. That’s 7 OOMs apart in parameters per token.
After a lot of painstaking work, we present new data-constrained scaling laws and learning algorithms to scale to (extrapolated) quadrillion-parameter transformers.
w/ @akshayvegesna, @EthanBaron78553, @asatheesh6561
Full paper led by @EthanBaron78553, with @ShmuelBerman, @andrewgwils
