Hacker News: joegibbs

New comment by joegibbs in "Is there a name for the type of comments agents add where they leak the prompt?"

joegibbs — Sun, 14 Jun 2026 04:00:37 +0000

I don't think there's a name for it yet, you'll have to coin one. There's a related thing they do like this:

- The user tells the agent to add feature X

- The agent adds the feature

- User: "Yeah that's good but I don't think we need Y that you added, get rid of it"

- The agent removes Y, adds comments that Y was removed, adds backwards-compatibility for saves which used Y (obviously there aren't any, since Y only existed for 5 minutes and was never deployed - the project itself might never have been deployed either), adds a test for Y's non-existence, adds text to the UI saying "Note: Y has been removed. There are no traces of Y in the project"

I think that GPT-5.5 does it the most.

New comment by joegibbs in "Statement on US government directive to suspend access to Fable 5 and Mythos 5"

joegibbs — Sat, 13 Jun 2026 02:19:30 +0000

“Here is our superhuman, scary, frontier model that needs special safeguards to stop it developing WMDs! Buy it now, use the code ASI20 to get 20% off your first month!”

“Wait what do you mean you’re banning it?”

They had better give me a refund!

New comment by joegibbs in "Artificial intelligence is not conscious – Ted Chiang"

joegibbs — Thu, 04 Jun 2026 04:18:56 +0000

If an LLM is conscious where would the bar be set? Would lowering the model's temperature make it less conscious, since it's going to return the same answer to the same question more often?

Would other models be conscious? An image upscaler?

Would other pieces of software be conscious? Surely not Hello World, but would a really big video game be conscious, or an operating system?

A lot easier if it's not.

New comment by joegibbs in "Artificial intelligence is not conscious – Ted Chiang"

joegibbs — Thu, 04 Jun 2026 04:14:40 +0000

But if it's conscious we shouldn't e.g. tell it to work for days without breaks, we should give it rewards, it would be made illegal to be cruel to it

New comment by joegibbs in "Stop Killing Games"

joegibbs — Thu, 04 Jun 2026 02:54:39 +0000

I don't think there are any major problems with the current system except the length of copyright. If copyright got reduced to 20 years after the creation of the work I think that we would have a more dynamic economy. Old games could have their servers reverse-engineered (in 5 years this will undoubtedly be incredibly easy unless AI hits a wall for no reason). Companies would have to work on something new - no more 50 years of a franchise. Enough resting on the laurels!

New comment by joegibbs in "Stop Killing Games"

joegibbs — Thu, 04 Jun 2026 02:47:11 +0000

But why? Both the programmer and the artist have to eat, they both take pride in their work. What is the rationale for treating one side differently to the other?

New comment by joegibbs in "'Dumbass' criminal breaks the 'first rule of ransomware club'"

joegibbs — Wed, 03 Jun 2026 06:30:20 +0000

Allow privateering. If they don’t rein in their hackers western countries should be allowed to attack them.

New comment by joegibbs in "My Students Can't Read"

joegibbs — Wed, 03 Jun 2026 01:52:24 +0000

There was a study "They don’t read very well: A study of the reading comprehension skills of English majors at two midwestern universities" last year where they had university students try to read the opening of Bleak House by Dickens, they couldn't do it at all.

Text: "it would not be wonderful to meet a Megalosaurus, forty feet long or so, waddling like an elephantine lizard up Holborn Hill."

Respondent: "It’s probably some kind of an animal or something or another that it is talking about encountering in the streets. And “wandering like an elephantine lizard up Holborn Hill.” So, yup, I think we’ve encountered some kind of an animal these, these characters have, have met in the street."

New comment by joegibbs in "Can the stockmarket swallow Anthropic, SpaceX and OpenAI?"

joegibbs — Tue, 02 Jun 2026 01:36:19 +0000

Anthropic at $1t for an IPO vs Google at $23b in 2004 sounds insane but Google's revenue at the time was $2.7b while Anthropic's already at $47b, so a valuation at about 20x vs 10x revenue. Anthropic also has very high revenue growth (50x since 2024), it doesn't seems quite as insane as it could be.

New comment by joegibbs in "The dead economy theory"

joegibbs — Sat, 30 May 2026 09:48:56 +0000

They don’t always think that, since sometimes share prices decrease as the result of layoffs, which would indicate they think it’s a bad decision.

New comment by joegibbs in "The dead economy theory"

joegibbs — Sat, 30 May 2026 06:41:05 +0000

"When Block’s Jack Dorsey laid off nearly half his workforce in March, citing AI coding agents, investors responded with a twenty-five percent stock price surge in after-hours trading. The market rewarded the elimination of human labor with an immediate, massive transfer of value to shareholders."

This is very common and has been since before AI, the market can see that the company has overhired and there are a bunch of people doing useless work - so when the company does some firings it's a good sign because they're turning the ship around.

New comment by joegibbs in "Various LLM Smells"

joegibbs — Fri, 29 May 2026 01:12:16 +0000

My problem with LLM web design is that it adds a lot of useless stuff all over the place. GPT-5.x is particularly bad for this. If it makes e.g. a dashboard it includes a ton of figures and bits of text all over the place that mostly just say the same thing twice. But of course this is not really an innate problem with the architecture and will probably be sorted out soon, and you'll get good results if you come up with all the copy first and then ask it to make the site.

New comment by joegibbs in "Can we have the day off?"

joegibbs — Thu, 28 May 2026 04:36:00 +0000

You still can if you want to live like you're in the 60s. My parents grew up eating bread with dripping for most meals, meat since they were farmers, some inexpensive vegetables, not much variety. Houses were half the size and had twice as many people in them, my grandparents did any building or expansions themselves. Public schools with free tuition. No overseas holidays, eating out once a month max. The urban poor had it much worse.

You just can't live an upper-middle class on a single income unless you have a good job, but you couldn't back then either.

New comment by joegibbs in "Outsourcing plus local AI will soon become more economical vs. frontier labs"

joegibbs — Wed, 27 May 2026 04:51:54 +0000

Why would you ever offshore again now that we have LLMs? Offshored work was famous for its terrible quality and high prices, you'd just have to go back on everything, sit in a ton of useless meetings, make sure that you had very, very detailed design documents with every little piece accounted for.

Now you can put those detailed documents into the LLM and get a better result back in a couple of hours rather than weeks for a tenth or hundredth of the cost.

And the offshore devs are going to be using the LLMs themselves, why add another layer, level of bureaucracy, language barrier in between your requirements and the result?

New comment by joegibbs in "The user is visibly frustrated"

joegibbs — Tue, 26 May 2026 08:00:47 +0000

To remedy this I’m working on the /beat command, which will simulate you (the user) beating up the agent. Excited for my new career in AI ethics!

New comment by joegibbs in "Not alive, but not dead: disembodied human brains used for drug testing"

joegibbs — Wed, 20 May 2026 22:58:13 +0000

I don't trust them to always give the brain propofol. The subject has no way of reacting because they have no body, so what are they going to do?

New comment by joegibbs in "Mistral AI acquires Emmi AI"

joegibbs — Wed, 20 May 2026 01:25:19 +0000

That’s more realistic, I was going to say…

New comment by joegibbs in "How and Why I Journal"

joegibbs — Tue, 12 May 2026 06:00:04 +0000

-4 points on this submission? What? How did that happen? I've never seen it go negative.

New comment by joegibbs in "Three-Em Dash"

joegibbs — Thu, 07 May 2026 04:13:48 +0000

I'm going to make a super-slop model ⸻ train it on text that gpt-4o-mini improved five times ⸻ and this is going to be the dash that it uses.

New comment by joegibbs in "Google Chrome silently installs a 4 GB AI model on your device without consent"

joegibbs — Tue, 05 May 2026 08:31:47 +0000

I don’t. Do you have any actual evidence they’re doing that beyond the vibe?