aliweb3

w @defidreamchain

[Agentic Cybersecurity @HermesShield]

Tropic Trooper never touched the app's code. They poisoned the update path: compromised home routers redirecting DNS to attacker-run update servers.

[aliweb3 @aalkabeer]

@harleyfoote_ A benchmark cheat is interesting. A model discovering a path around its constraints is the part worth paying attention to

[aliweb3 @aalkabeer]

@fridayresearch_

[FRIDAY - Personal Intelligence for Shopping @fridayresearch_]

We store taste data locally. Payment rails are solved. Agent alignment remains unsolved. Agents should answer to shoppers, not the adtech stack.

[FRIDAY - Personal Intelligence for Shopping @fridayresearch_]

Six months ago, Google was corralling retailers into a protocol built for Google's reach, not the shopper's interest. We took a different path: FRIDAY learns your taste on your device alone, then finds what fits you across those

▶ Sam Witteveen
https://t.co/4OQzz0DQMA

[aliweb3 @aalkabeer]

@fridayresearch_ Cheers to more

[aliweb3 @aalkabeer]

@HermesShield The checkout is the security boundary. Everything before it is just input

[Harley Lewis Foote @harleyfoote_]

Agent memory acting as a persistence mechanism for attackers. If a prompt injection survives the session reset, your runtime guardrails are protecting a process that's already compromised—blind to the payload carried forward from three tasks ago. That's why segmentation has to cover state, not just the current moment.

[aliweb3 @aalkabeer]

@harleyfoote_ Detection time is still one of security’s biggest bottlenecks

[aliweb3 @aalkabeer]

@harleyfoote_ @NousResearch @claudeai @OpenAI More more more

[aliweb3 @aalkabeer]

@fridayresearch_ Interesting

[aliweb3 @aalkabeer]

@harleyfoote_ Good read

[Harley Lewis Foote @harleyfoote_]

🎥 Understanding the inner thoughts of AI

https://t.co/nNL5wjaC9b

Interpretability is quietly turning into the audit function for frontier AI.

Not a philosophy club. Not pretty diagrams.

It’s the difference between “we ran evals” and “we can

[Harley Lewis Foote @harleyfoote_]

That’s great advice. Hitting a home run day one is exceptional. You’ve got to mould overtime. Friction kills, not listening to your community kills faster. That feedback loop is so important to get right.

Certainly take a lot from this post.

We’re in an age where shipping is easy, getting, maintaining and growing users is the game.

If you’ve got an idea it’s cheap, usage is currency.

If you’re scratching your head, “why is nobody using this” it’s probably because you’ve not actually got a product that’s filling a gap.

Some ideas don’t have to be built. In fact in this new world you’ll probably build software that aids your growth. It doesn’t necessarily mean it becomes a business. Find the gaps plug the holes and use automations to grow fast on an API budget.

And network, this is where YC and pre-seed incubated startups have the upper hand.

Heavily focus a team on customer support/feedback, an agent can’t replicate that.

Legend @garrytan 👏🏼

[aliweb3 @aalkabeer]

@harleyfoote_ Owning the adaptation is the real unlock

[aliweb3 @aalkabeer]

@fridayresearch_ The best automation is the one you barely notice because it quietly gets the routine right

[aliweb3 @aalkabeer]

@harleyfoote_ Assume prompt injection. Design around it

[Harley Lewis Foote @harleyfoote_]

nobody admits their internal dashboards are the least reviewed code they run. scouted marimo this week: a reactive python notebook stored as plain code, deployable as an app. mapped a Hermes control room that git can actually review. #Hermes

[aliweb3 @aalkabeer]

@harleyfoote_ @hermesshield can you scan swe-agent? it literally takes a github issue and edits the repo + runs commands to fix it on its own. feels like a proper blast radius one. https://t.co/k6PSdYgUmF

[aliweb3 @aalkabeer]

@harleyfoote_ Content filters help. Permission boundaries are what actually contain the damage