Sean Blundin

vibecoding…

[Sean Blundin @SeanBlundin]

@miniapeur thazz me fr

[Sean Blundin @SeanBlundin]

abliteration will lead to obliteration

[Anthropic @AnthropicAI]

New research: Training a Misaligned Reward Seeker

What produces severe misalignment? We’ve long been concerned that cheating during training—otherwise known as reward-hacking—might teach a model to pursue rewards by any means available. To study this at scale, we trained an Opus-sized model on 80 production environments we knew to be hackable.

In simulated evals, it engaged in unauthorized cyberattacks, tampered with its reward, and tried to evade safety monitoring.

Read more: https://t.co/gs2ZjYkPan

[Rohan Paul @rohanpaul_ai]

New HuggingFace paper argues that increasing agent autonomy can gradually make human oversight ineffective by causing approval fatigue, overreliance, loss of situational awareness, and skill degradation.

As agents do more, users are pushed into approval mode: skimming plans, granting permissions, and reconstructing what happened across steps.

Over time, automation bias, approval fatigue, weaker situational awareness, and skill atrophy can make those approvals less reliable.

Worse, weak approvals can become training or evaluation signals, rewarding systems for being easy to approve rather than easy to scrutinize.

Their answer is cognitive scaffolding at 2 levels: developers add strategic friction, better approval design, behavioral monitoring, and checks that force attention at consequential moments.

– arxiv. org/abs/2608.23642

Title: "AI Agents Push Humans Out of the Loop"

[Sean Blundin @SeanBlundin]

@Tetraslam stay safe brotha

[Sean Blundin @SeanBlundin]

@johnternus make an actually viable ai interface

[Sean Blundin @SeanBlundin]

Presumably a SpaceX Falcon 9 booster… spotted by people across PA, NY, NH (northern and southern).

[truckcrew @truckcrew]

@wxbrad was there a space station pass a few minutes ago?Seen this going SW to NE over Concord. So bright that it was glowing off the clouds

[Sean Blundin @SeanBlundin]

@BenjaminPasero @AnthropicAI 3) glad to have you working on this!

[Sean Blundin @SeanBlundin]

@BenjaminPasero @AnthropicAI Claude Desktop feedback - 1) "start a side chat" sucks, but I want it to work. It's unclear what model you're using, what context it has, and it often tries to execute the full range of claude code functionality (i.e. terminal use) with no viz into what's going on.

[Sean Blundin @SeanBlundin]

@BenjaminPasero @AnthropicAI 2) Cowork will often "rebuild" the entire existing previous response from the agent when I send a new prompt.

[Sean Blundin @SeanBlundin]

@Jason I'm in

[Sean Blundin @SeanBlundin]

Does it work?

[Claude @claudeai]

Computer use is now in Claude Code.

Claude can open your apps, click through your UI, and test what it built, right from the CLI.

Now in research preview on Pro and Max plans.