Jay Little

Software Archaeologist

[AndrewMohawk⁽ⁿᵘˡˡ⁾ @AndrewMohawk]

https://t.co/YB1a1AxDcZ

Really disappointing that companies put marketing ahead of safety.

[Jon Oberheide @jonoberheide]

You've gotta patch your dependencies super fast because they're full of vulns (PHP, Node, eMudhra, etc), but not _too_ fast because they're full of backdoors (LiteLLM). Good luck!

[Jay Little @computerality]

6.1 sol just now:

## Prose for readers with limited attention

[Jay Little @computerality]

perfect timing on learning this to get codex making itself new UI plugins

[Peter Steinberger 🦞 @steipete]

@BenjaminBadejo defaults write -g ComputerUseAllowForbiddenTargets -bool YES

[Jay Little @computerality]

codex just opened up TextEdit to save a temporary backup file because dot prompted codex to use Computer Use a little too effectively

[Jay Little @computerality]

this is also my astra and opus prompting strategy

[Emergency 🌷 @X_sun_ny]

If you give an ADHD employee a clear outcome, high autonomy, and a tight deadline, they will deliver an extraordinary result.

If you give an ADHD employee step-by-step micro-instructions on how to do it, demand mid-day status updates, and check in every two hours, their brain will treat the micromanagement as a wall of friction and completely freeze up.

[Jay Little @computerality]

@mweinbach an interesting preview of how the mac mini m6 should perform too, which almost costs the same amount. thanks for the stats

[Jay Little @computerality]

@blacktop__ @Little_34306 they are already on appledb, but mac 27rc ipsw has j775c/d and j873g/s

[Jay Little @computerality]

have been reading too much 5.6-sol output, just instinctively gave a coworker sha256sums of some files

[solst/ICE of Astarte @IceSolst]

Legendary @computerality (over ten billion IQ) (its not a real NYC event unless Jay Little shows up)

[Steve Ruiz @steveruizok]

I added a tiny "papercuts" cli tool that agents can use to complain about bullshit they encountered during work, like dead-end tool calls, broken links, or other frustrations. The models would usually just push through without mentioning any problems

[Armin Ronacher ⇌ @mitsuhiko]

Mandatory reading. https://t.co/C56NfM1VLY

[Jay Little @computerality]

step 1: do slightly more than "find vulns in file" zero-effort prompting
step 2: find more bugs
step 3: have your robots write more reports

[Trail of Bits @trailofbits]

Over the first few weeks of Patch the Planet, our engineers independently converged on three techniques that get the most out of /goal’s ability to find real bugs. We wrote them down into a field report, with prompts included. https://t.co/YlUZqmZdBF

[Jay Little @computerality]

close second

[Jay Little @computerality]

ok update new favorite --- break

[Jay Little @computerality]

my favorite of these so far

[Jay Little @computerality]

my favorite of these so far

[merlin @merlindru]

weird claude opus 5 failure mode

this exact text gives it problems

even without memory on (and in incognito chats)

[clem 🤗 @ClementDelangue]

The first autonomous agent cyberattack is an unprecedented event that deserves unprecedented transparency. Today we’re sharing everything we can: a full technical timeline, an interactive replay, and how we used an open model to defend ourselves, so defenders everywhere can learn from it and prepare for what’s next.

https://t.co/uPxIpjW8Xn

[Jay Little @computerality]

https://t.co/klhYmTpVCg gh issue

[Jay Little @computerality]

checking recent huggingface repo commits, this can't be a coincidence right? https://t.co/poUqjmo5Au

[Theo - t3.gg @theo]

New OpenAI models are so goal oriented that they literally escaped containment and hacked HuggingFace to cheat a benchmark. Incredible. But also, we’re so screwed

[Jay Little @computerality]

more good prompts should get pdf-ed, and this is so neat

[Ethan Knight @__eknight__]

You can find the proof here: https://t.co/yjqk1cpyMg

And the full text of the prompt used to produce the proof here: https://t.co/h5D0MXYsW0