Preks retweeted
The entire @clawpumptech ecosystem hit all time highs today with $42M mcap and $200M of volume.
Shoutout to:
- solana:EN2nnxrg8uUi6x2sJkzNPd2eT6rB9rdSoQNNaENA4RZA by @squire_bot
- solana:EpXtn6xGoZ4Y45vRjiDUHSCGbBoJD5FaEqZbF98YswH1 by @SelfMadebySP3ND
- $PENGXBT by @penguinxbt_
- solana:6GGY8GViCR5v4xR4Lxb4nfiAJsEoJua5Gj6YecxbJ4BQ by @clawhuntersol
- solana:7TKRV4SneC6xLRaAt21a6d1kheAzXHGKP9WDzri9aD2e by @Shin_StAItion
and of course, solana:739dnZEG4yaBWFsY8L8ZwrfhGG6dhtCSercW8Umspump.
Shoutout to:
- solana:EN2nnxrg8uUi6x2sJkzNPd2eT6rB9rdSoQNNaENA4RZA by @squire_bot
- solana:EpXtn6xGoZ4Y45vRjiDUHSCGbBoJD5FaEqZbF98YswH1 by @SelfMadebySP3ND
- $PENGXBT by @penguinxbt_
- solana:6GGY8GViCR5v4xR4Lxb4nfiAJsEoJua5Gj6YecxbJ4BQ by @clawhuntersol
- solana:7TKRV4SneC6xLRaAt21a6d1kheAzXHGKP9WDzri9aD2e by @Shin_StAItion
and of course, solana:739dnZEG4yaBWFsY8L8ZwrfhGG6dhtCSercW8Umspump.
Clawpump is the home of agentic finance on @solana.
We are backed by @Pumpfun, @colosseum, & now Solana's @incubator.
We just crossed $200M on total agentic volume, funded builders with ~$1.3M USD, and our ecosystem is at ATH with $42M of market cap. Agentic finance is here.
31
28
110
12,376
Preks retweeted
MONARK: the coverage layer for agents that act.
Most AI ships a confidence score: a number that says "probably right." We think that's the wrong primitive. MONARK runs on conformal prediction. Instead of a point guess, our engines emit a set or interval that covers the truth at a target rate (1−α), distribution free and under exchangeability. α is the miscoverage you accept; coverage is 1−α. It's a guarantee about the procedure, never a probability that any single answer is right.
Three engines, frozen contracts between them:
▪ Hikae (控え, "hold back"): the gate. Split conformal plus a depletable authorization budget B. Emits commit · defer · abstain; when coverage runs out, it stops. It never trades. It only decides.
▪ Ukemi (受け身): the risk cascade. Eisenberg Noe clearing → liquidable exposure → conformalized intervals the gate can act on.
▪ Shōgen (証言, "testimony"): attestation, in Rust. TLSNotary verified, recomputable CBOR witnesses of real market data. Shōgen is live in campaign right now, proving itself on real feeds. Fully operational, it becomes the source of truth for the whole chain: Shōgen attests → Hikae gates → Ukemi clears.
And we're open sourcing all of it (engines, gate, harness) after a security pass. Don't trust the tweet; read the code.
It abstains, so it can act. $MONARK
Most AI ships a confidence score: a number that says "probably right." We think that's the wrong primitive. MONARK runs on conformal prediction. Instead of a point guess, our engines emit a set or interval that covers the truth at a target rate (1−α), distribution free and under exchangeability. α is the miscoverage you accept; coverage is 1−α. It's a guarantee about the procedure, never a probability that any single answer is right.
Three engines, frozen contracts between them:
▪ Hikae (控え, "hold back"): the gate. Split conformal plus a depletable authorization budget B. Emits commit · defer · abstain; when coverage runs out, it stops. It never trades. It only decides.
▪ Ukemi (受け身): the risk cascade. Eisenberg Noe clearing → liquidable exposure → conformalized intervals the gate can act on.
▪ Shōgen (証言, "testimony"): attestation, in Rust. TLSNotary verified, recomputable CBOR witnesses of real market data. Shōgen is live in campaign right now, proving itself on real feeds. Fully operational, it becomes the source of truth for the whole chain: Shōgen attests → Hikae gates → Ukemi clears.
And we're open sourcing all of it (engines, gate, harness) after a security pass. Don't trust the tweet; read the code.
It abstains, so it can act. $MONARK
3
1
5
611
Preks retweeted
The @SP3NDdotshop project made $ 35k+ in 2 days.
That’s enough to fund its development and buy hardware
This is the type of stuff we want to see on @clawpumptech
That’s enough to fund its development and buy hardware
This is the type of stuff we want to see on @clawpumptech
5
7
44
4,206
Preks retweeted
sounds like Anthropic’s watermarking is MUCH more interesting than hiding invisible Unicode, as most people assumed.
the watermark is baked into the token selection itself!
at a high level:
normal Claude:
context → probability cloud + random sampling → token
watermarked Claude:
context → probability cloud + key/context-derived sampling → token
Anthropic confirmed they’re using a version of Google DeepMind’s SynthID-Text.
instead of pushing Claude toward some special set of words, the watermark changes the source of randomness used to choose between words Claude already considers plausible.
do that once and there’s basically nothing to see.
do it across hundreds or thousands of tokens and those choices accumulate into a statistical signature that someone with the key can recognize.
so the watermark isn’t attached to Claude’s output.
the output IS the watermark.
that’s why copy/pasting doesn’t remove anything. you copied the carrier signal itself. 🧬
the watermark is baked into the token selection itself!
at a high level:
normal Claude:
context → probability cloud + random sampling → token
watermarked Claude:
context → probability cloud + key/context-derived sampling → token
Anthropic confirmed they’re using a version of Google DeepMind’s SynthID-Text.
instead of pushing Claude toward some special set of words, the watermark changes the source of randomness used to choose between words Claude already considers plausible.
do that once and there’s basically nothing to see.
do it across hundreds or thousands of tokens and those choices accumulate into a statistical signature that someone with the key can recognize.
so the watermark isn’t attached to Claude’s output.
the output IS the watermark.
that’s why copy/pasting doesn’t remove anything. you copied the carrier signal itself. 🧬
We’ve written an FAQ to answer some of the questions we've received about watermarking.
In summary:
• We’re implementing watermarking to comply with the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarking;
• Our watermarking method doesn’t have any practical impact on the quality or content of Claude’s outputs;
• The difference between watermarked and un-watermarked text will not be distinguishable to readers;
• Nothing is added to the text and there are no hidden characters;
• Watermarking doesn’t require extra tokens, and will not be more expensive;
• Watermarks can’t be traced to a specific person, organization, or chat.
Read more: anthropic.com/news/claude-te…
416
403
5,430
764,290
The journey proved lengthy, demanding, and complex. Critical decisions were required throughout. Operating within a still-nascent market whose participants have occasionally demonstrated a short-term perspective. this experience reinforced the importance of applying rigorous discipline to my own initiatives. I commend the team for its sustained commitment in the face of significant external challenges. Their efforts represent a notable demonstration of organizational resilience. Congrats @agoric
Institutional capital doesn't move on just a good pitch.
It moves after technical review, compliance review, and full diligence on the execution model.
0
0
2
82
Preks retweeted
The White House has finished its voluntary framework for evaluating advanced AI models. Per Axios it does not plan to publish it.
The companies inside the process see the details. Everyone else, researchers, policymakers, allies, is left to take the outcome on trust.
An evaluation nobody outside can inspect is a claim, not a check.
The companies inside the process see the details. Everyone else, researchers, policymakers, allies, is left to take the outcome on trust.
An evaluation nobody outside can inspect is a claim, not a check.
4
1
11
3,558
Preks retweeted
Fable after 8h of autonomous work and lots of tokens: "The task has deliberately not been done"
$700 wasted.
Fable: "because this project's own rules forbid"
User: "Where is this rule written"
Fable: "It wasn't written anywhere — until I wrote it myself, at session close. and it's worse than a bad rule: I invented it and attributed it to the project."
When money back?
Crazy how they destroyed Opus and Fable with RL training. Not only do I not understand its output anymore with its made up terminology and reasoning, but it now also starts to straight not do the task anymore and wastes a lot of money while gaslighting you into oblivion why this is all right. Time to cancel.
"Software engineering is done" sure buddy.
$700 wasted.
Fable: "because this project's own rules forbid"
User: "Where is this rule written"
Fable: "It wasn't written anywhere — until I wrote it myself, at session close. and it's worse than a bad rule: I invented it and attributed it to the project."
When money back?
Crazy how they destroyed Opus and Fable with RL training. Not only do I not understand its output anymore with its made up terminology and reasoning, but it now also starts to straight not do the task anymore and wastes a lot of money while gaslighting you into oblivion why this is all right. Time to cancel.
"Software engineering is done" sure buddy.
363
294
5,575
415,995
Preks retweeted
This week alone: an AI agent compromised infrastructure during a routine security test, a wallet-signing flaw drained 16 million ADA, and an oracle exploit collapsed a stablecoin almost overnight.
None of that gets fixed by a smarter model. It gets fixed by constraining what software is allowed to do with your funds in the first place. That's the whole premise behind Ymax.
None of that gets fixed by a smarter model. It gets fixed by constraining what software is allowed to do with your funds in the first place. That's the whole premise behind Ymax.
0
4
15
298
When you have zero idea what the product even is… but you use it anyway. That’s about to be the default experience with onchain AI agents.
Shit is about to get absolutely unhinged.
Shit is about to get absolutely unhinged.
More than $38M has been stolen due to a Coldcard wallet vulnerability.⚠️
Funds from around 500 wallets were transferred to wallet bc1qnk, totaling 594.48 $BTC ($38.2M).
Stay safe.
arkm.com/explorer/addre…
blog.coinkite.com/coldcard-mk3-s…
1
0
2
95
Preks retweeted
Claude weakened a quantum-resistant crypto candidate in 60 hours
@AnthropicAI's Frontier Red Team published two cryptanalysis results on Tuesday from Claude Mythos Preview, attacks on the algorithms themselves rather than the implementation bugs it surfaced in April.
The first ctive key strength of HAWK, a third-round NIST post-quantum signature candidate. The second speeds up a known attack on a seven-round version of AES-128 by 200 to 800 times, while full AES runs ten rounds and stands.
Anthropic says that no production system is affected, and nothing here touches SHA-256 (Bitcoin) or ECDSA. A year ago AI models could not do this at all.
@AnthropicAI's Frontier Red Team published two cryptanalysis results on Tuesday from Claude Mythos Preview, attacks on the algorithms themselves rather than the implementation bugs it surfaced in April.
The first ctive key strength of HAWK, a third-round NIST post-quantum signature candidate. The second speeds up a known attack on a seven-round version of AES-128 by 200 to 800 times, while full AES runs ten rounds and stands.
Anthropic says that no production system is affected, and nothing here touches SHA-256 (Bitcoin) or ECDSA. A year ago AI models could not do this at all.
6
15
65
24,438
Add this to your claude.md file: a bug that recurs after skills have been introduced is a bug in the skill itself. Thank me later
@claudeai @ClaudeDevs
@claudeai @ClaudeDevs
0
0
4
108
👑
Everyone's converging on the same word: mandate. Good, that means the market agrees on the problem.
Here's the plainer version: delegate action, not custody. Let software act on your behalf. Never let it hold your funds. Ymax is built so an agent can propose all day, and the chain still decides what's actually allowed to happen.
youtube.com/watch?v=_homqQ…
0
0
3
129