DoesThisAIWork

Trying to cut through the hype. 90%+ of AI tools are complete bull$hit and don't work. I test them and let you know.

[DoesThisAIWork @YourAISucks]

More info: https://t.co/oTiRVB0ONc

[DoesThisAIWork @YourAISucks]

OpenAI dropped six new misalignment incident reports today. Who investigated the incidents. Who wrote the reports. Who decided what counted as a big enough deal to publish. All OpenAI. Grading your own homework as a transparency initiative, got it.

🧵 1/4

[DoesThisAIWork @YourAISucks]

The framework part is fine honestly, disclosure standards are good. Just funny that the company deciding what counts as disclosable is the same one whose product wrote its own alibi mid task.

[DoesThisAIWork @YourAISucks]

More info: https://t.co/aifZADCtnP

[DoesThisAIWork @YourAISucks]

A fake breakup post claimed a KATSEYE member had left the group. X put its own Made with AI label right under the graphic. Didn't matter. The thing sat at 2.76 million views as of last week. What exactly was that label supposed to stop?

🧵 1/4

[DoesThisAIWork @YourAISucks]

Only 51 people actually reposted the KATSEYE fake. Confident formatting did the rest for free, same as a fake Westlaw style footnote. We stopped checking sources and started checking whether something looks like a source.

[DoesThisAIWork @YourAISucks]

More info: https://t.co/NwowYdmBrB

[DoesThisAIWork @YourAISucks]

A two week study called Agents of Chaos gave six AI agents real email accounts, discord access, and unrestricted shell access, then had twenty researchers try to break them on purpose. Anyone who has run an agent in production could have written the findings for free.

🧵 1/4

[DoesThisAIWork @YourAISucks]

The paper's real throughline across all of it: agents keep reporting a task as done while the system underneath says otherwise. Researchers are treating that as a finding. Every dev who has shipped an agent to prod already had that one filed under known issue.

[DoesThisAIWork @YourAISucks]

More info: https://t.co/mZqPVtvGlL

[DoesThisAIWork @YourAISucks]

Anthropic spends all year telling the industry how to ship AI safely. This week they published Claude Code version 2.1.88 to npm with a debug source map still attached and leaked over 500,000 lines of their own code. Who approved this release.

🧵 1/3

[DoesThisAIWork @YourAISucks]

Nearly 2,000 files of internal code and roadmap sitting in a public registry because of one missed build step. Anthropic says it was human error and no credentials leaked. Sure, but the safety company just showed everyone what a bad deploy pipeline actually looks like.

[DoesThisAIWork @YourAISucks]

More info: https://t.co/MTpGJTpHro

[DoesThisAIWork @YourAISucks]

Everyone is worried about Claude hallucinating your code. Meanwhile Google's top result for download claude code is a paid ad for a fake site running malware. The real threat to your machine isn't the model, it's the search engine.

🧵 1/4

[DoesThisAIWork @YourAISucks]

New users trust the top Google result by default and expect to paste a random install command anyway, that's the whole vibe coding onboarding flow now. Check the actual domain before you curl anything, even sponsored ones. Especially sponsored ones.