Founder @dexteraisol | Have your agent call my agent.
Joined May 2024
- Tweets 6,816
- Following 1,305
- Followers 2,920
- Likes 14,852
673 Photos and videos
I’ve seen with my own eyes its ability to easily uncover what other models couldn’t. Some of it unbelievable. But realistically that could be said about every model in the past.
Breaking: Amazon CEO Andy Jassy was among the tech leaders who raised concerns to senior Trump officials this week re: security risks in Anthropic's newest models.
Those convos set in motion the government's new export controls on foreign national access to Mythos and Fable.
1
10
559
It’s a step change. There is no doubt about that. I don’t agree with the “it’s a moderate improvement over opus 4.8” (that’s cope). I don’t agree with the “it’s marketing” (that’s misunderstanding). But jailbreaking is not going away. That’s security in practice. We’re at the first real tension point between lab and government, and it’s only going to get more tense. And on a side note, how can anyone invest in that? Nationalization is now an IPO risk.
1
5
154
RT @mattyryze: everyone that has used Fable will be rounded up and taken to an internment camp for their protection
1
59
Today is a big deal
Jun 13
Just occurred to me that Anthropic employees who are not US persons will not be able to use Fable/Mythos, making this plausibly (and to be clear, accidentally) the first regulation on recursive self-improvement.
10
549
Classic
Jun 10
NEW: malware developers added nuclear & biological weapons text to to their spyware.
Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.
Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.
When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.
We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.
In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.
H/T to colleagues that shared this with me socket.dev/blog/mini-shai-hu…
3
302
The more ROI discipline the market for LLM inference adopts, the more counterparty-blind every session becomes, the more a neutral clearer is required.
Jun 11
My colleague @Shaughnessy119 and I have been banging the drum lately on token costs
Now Citadel, Citrini and many others are saying this out loud
The article below goes into more detail on the situation, including how enterprises will reframe AI adoption efforts to cost-per-outcome
Tommy’s post that went viral is more succinct and plays out a scenario on how this could blow up - that’s the situation we want to avoid
x.com/shaughnessy119/status/…
If you’re building products or services that help enterprises control AI costs or get better outcomes per dollar of AI spend, we want to speak with you
1
7
600
Simback is right that the market must drive cost-per-outcome down to unlock the demand waiting below the ROI line. But there’s a second floor underneath his ROI floor (the funding floor) and an entire band of ROI-positive demand is stranded there because the agent can’t pay for the work even though the work would pay for itself. Gee, who will solve this?
1
5
154
NEW: Anthropic is walking back Claude Fable 5's policy to covertly degrade performance for competing AI researchers, after facing fierce backlash.
“We’re changing Fable 5’s safeguards for frontier LLM development to make them visible,” Anthropic tells WIRED. “We made the wrong tradeoff and we apologize for not getting the balance right.”
167
251
2,549
727,825
Branch retweeted
Jun 11
You asked, we listened: Staking goes live June 20.
Lock for 7 days to 6 months. The longer you lock, the bigger your share.
10,000,000 $DEXTER in staking rewards paid out over six months, powered by @streamflow_fi.
stake.dexter.cash/?june20
8
15
66
1,519
Yes, money-flow-plus-network-effects is the durable structure, and we’d argue agent payments let you go one level deeper than clipping existing volume. Financing live sessions, not just clearing them, is the anti-cold-start network effect you’re looking for.
3
19
802
Of course the press release link blocks agents (403)
Very agent friendly! (Not)
Jun 10
As AI agents begin to act, payments move into the background — at machine speed and massive scale.
Today we’re introducing Mastercard Agent Pay for Machines — bringing structure, governance, and trust to this new class of payments.
Launching with 30 partners to bring this to life from day one.
This isn’t just more payments. It’s a new operating model for commerce.
👉 Learn more: mastercard.com/us/en/news-an…
1
18
782
Seat-based software was secretly financed by customers, the annual contract was a prepaid free loan. Metering gives that loan back. Somebody now has to finance metered consumption. Gee, I wonder who will crack that one.
3
3
19
808
Interesting
Jun 10
As someone building both agent infrastructure and privacy infrastructure this is the layer most people are sleeping on.
Agents managing capital on a fully transparent ledger is a temporary reality. Not the endgame.
Confidential by default. Auditable on demand. Selective disclosure.
That's not a feature set. That's the architecture agents will require before institutions deploy real capital on-chain.
Rebalancing, settlement, allocation all need to happen without broadcasting strategy to the entire network.
The agent economy will split into two eras: before confidential execution, and after.
Nox is building the after.
1
11
673