Hacker News: david_shi

New comment by david_shi in "Samurai City"

david_shi — Thu, 04 Jun 2026 23:25:51 +0000

Perhaps Golden Ages are the rare and illusive times when a third way is possible.

New comment by david_shi in "Chipotlai Max"

david_shi — Tue, 02 Jun 2026 08:22:57 +0000

This is the singularity we were promised

Ask HN: How do you solve AI's confused deputy problem?

david_shi — Mon, 01 Jun 2026 03:40:23 +0000

An agent's value is proportional to the permissions it's been granted.

There's been a lot of hype around solutions like default denial proxies, key vaults, and more, but nothing seems to address the core tension: an agent can be tricked into doing an attacker's bidding.

The best thing I could think of was to just run an observer loop and monitor everything the agent does with another LLM, but I'm curious if anyone has an elegant solution.

Comments URL: https://news.ycombinator.com/item?id=48352369

Points: 2

# Comments: 1

New comment by david_shi in "MCP is dead?"

david_shi — Sat, 30 May 2026 07:23:52 +0000

A bit off topic, but I think Google's A2A protocol could be a sleeper hit vs. the MCP protocol.

Not because it's better, but with one switch a significant portion of web traffic can be directed to A2A servers through Google's new search box.

New comment by david_shi in "SF startup is testing robots in Airbnbs, and trashing them, lawsuit claims"

david_shi — Fri, 29 May 2026 04:11:39 +0000

which itself seems to have been just a ruse:

https://youtube.com/watch?v=QTcL6Xc_eMM

New comment by david_shi in "Amazon scraps AI leaderboard to stop workers chasing usage scores"

david_shi — Fri, 29 May 2026 02:30:17 +0000

Incredible how often perverse incentives are implemented, it's like there's no foresight or understanding of behavior at all.

New comment by david_shi in "Various LLM Smells"

david_shi — Fri, 29 May 2026 02:18:41 +0000

I've found double and triple negations ("Not this. Not that. Not that either.") instead of just removing stuff to be one of the clearest tells.

New comment by david_shi in "Various LLM Smells"

david_shi — Fri, 29 May 2026 02:17:14 +0000

"That holds."

New comment by david_shi in "Show HN: Hallucinate – Massively Multiplayer Online Rave"

david_shi — Fri, 29 May 2026 02:09:34 +0000

Any plans to add global chat?

New comment by david_shi in "Pope Leo XIV’s first encyclical Magnifica humanitas to be published May 25"

david_shi — Tue, 19 May 2026 01:47:55 +0000

It's interesting how natural historic mimesis seems to be in these vaunted roles.

Presidents have their favorite past counterparts, so did emperors, and clearly the Pope does as well.

Does this kind of imitation prevent truly creative action taking? Did Akhenaten have someone in mind when he declared his own religion?

New comment by david_shi in "Two EA-18 fighter jets collide at Mountain Home airshow, pilots ejected safely"

david_shi — Mon, 18 May 2026 03:34:57 +0000

When you have exorbitant privilege you tend to use it.

New comment by david_shi in "I tried to make Claude make me money on open-source bounties"

david_shi — Sat, 16 May 2026 23:10:36 +0000

Bounties are the most overfished pond when trying to making money with AI because the task to earnings step is atomic.

AI UGC with affiliate sales and other kinds of monetized AI content creation requires niche selection, monitoring, and some upfront risk, but has much better ROI on effort.

New comment by david_shi in "Show HN: Race to the Bottom"

david_shi — Fri, 15 May 2026 01:58:46 +0000

What technology isn't surveillance technology these days?

New comment by david_shi in "New arXiv policy: 1-year ban for hallucinated references"

david_shi — Fri, 15 May 2026 01:48:46 +0000

In that case, you would just not do a reference. End to end autonomous science might have fewer concrete citations as the contributing knowledge is just the sum of the training data of the model.

New comment by david_shi in "Agents can now create Cloudflare accounts, buy domains, and deploy"

david_shi — Wed, 06 May 2026 17:51:36 +0000

I have to stop myself from appending "autonomous" any time I read agents.

Cursor and Claude Code are also agents, and making it easier to run these operations from those interfaces assuming some margin of error is ok is a genuinely useful feature.

New comment by david_shi in "Mike: open-source legal AI"

david_shi — Thu, 30 Apr 2026 22:39:58 +0000

I'd imagine it's like using Cursor/Claude Code vs. a Jetbrains IDE plugin.

New comment by david_shi in "A playable DOOM MCP app"

david_shi — Wed, 29 Apr 2026 00:42:03 +0000

what interfaces would this work on?

in-app for chatgpt, TUIs (https://www.youtube.com/watch?v=fdbXNWkpPMY, from another comment in the thread), anything else?

New comment by david_shi in "Microsoft and OpenAI end their exclusive and revenue-sharing deal"

david_shi — Mon, 27 Apr 2026 22:43:01 +0000

What aspects of the deal do you think kneecapped OpenAI the most?

New comment by david_shi in "The Prompt API"

david_shi — Mon, 27 Apr 2026 21:26:23 +0000

Will this API and others like it will be a strong enough incentive to move away from Chromium based browsers and back on to Chrome?

New comment by david_shi in "The Stanford Freshmen Who Want to Rule the World"

david_shi — Sat, 25 Apr 2026 20:40:51 +0000

Many such cases