gwerbin

13 ms·

271.
by gwerbin 6mo ago
Revenue-positive bugs are the stickiest features.
272.
by gwerbin 6mo ago
The thing that seems to bring up these extremely unlikely destructive token sequences and it totally seems to be letting agents just run for a long time. I wonder if some kind of weird subliminal chaos signal develops in the context when th
273.
by gwerbin 6mo ago
The author definitely deserves a lot of blame here and clearly doesn't understand AI well enough to have a coherent opinion on AI safety. But Railway bears some responsibility too because, at least of the author is to be believed, it l
274.
by gwerbin 6mo ago
Yeah that's the typical junior engineer scenario right? Run a command that wasn't meant to be destructive but accidentally destroy something. This is different. AI agent went on some kind of wild goose chase of fixing problems, an
275.
by gwerbin 6mo ago
But at least you have a 5000 LoC project on Github that deletes LinkedIn profiles!
276.
by gwerbin 6mo ago
Call me crazy but does AI not seem like the root cause here? At the beginning of the post they say that the AI agent found a file with what they thought was a narrowly scoped API token, and they very clearly state that they never would have
277.
by gwerbin 6mo ago
It's source-available proprietary software that happens to be distributed through NPM.
278.
by gwerbin 6mo ago
That's a common phenomenon in model fitting, depending on the type of model. In both old school regression and neural networks, the fitted model does not distinguish between specific training examples and other inputs. So specific inpu
279.
by gwerbin 6mo ago
This is just goofy prompting. I have good success when I ask the agent to help me debug the harness. "Help me debug why Claude Code is ignoring my hook".
280.
by gwerbin 6mo ago
That's the point. Burgers are more expensive (relative to "all" other goods) compared to back then.
281.
by gwerbin 6mo ago
Or just don't use AI to write code. Use it as a code reviewer assistant along with your usual test-lint development cycle. Use it to help evaluate 3rd party libraries faster. Use it to research new topics. Use it to help draft RFCs and
282.
by gwerbin 6mo ago
Sorta? Python has fairly strong types but it's no fun debugging a `None has no attribute foo` error deep inside some library function with a call site 1000 LoC away from the actual place where the erroneous None originally arose, due t
283.
by gwerbin 6mo ago
Nim type inference was a joy to use although I haven't touched the language in several years due to the language community seeming to collapse a bit.
284.
by gwerbin 6mo ago
What setup are you using? What models, what hardware, what agent harness, etc? I have the vague sense that this is all possible right now, but the amount of tinkering required doesn't seem worth it compared to, like, just not using A
285.
by gwerbin 6mo ago
It's less about the company leaving the stock market and more about "Private Equity" often being a legalized embezzlement scam designed to suck the company dry and then dump its withered husk in bankruptcy court.
286.
by gwerbin 6mo ago
That's not a bug, that's a feature
287.
by gwerbin 6mo ago
And they can offer an age verification product now too.
288.
by gwerbin 6mo ago
Which incidentally is the shuttle that brought back LDEF.
289.
by gwerbin 6mo ago
> And what percentage of a house's price is the building? Anecdotally like half, depending on the area. Plots of land go for $500k in Boston suburbs and new construction homes go for $1m.
290.
by gwerbin 6mo ago
F# has units, does it not support bounds on those units?
291.
by gwerbin 6mo ago
Let's say I'm an industrial monitoring SaaS company with a bunch of analytics products and a new AI intelligence product. Forget tracking workers' movements and stuff like that because that's even more complicated (the
292.
by gwerbin 6mo ago
> Plus, isn't the appeal of LLMs broadly that they can do somewhat-useful things with mostly-arbitrary input (if you ignore the risk of prompt injection)? They can definitely read HTML, but they do better with more structure. I prop
293.
by gwerbin 6mo ago
I'm also not advocating for or against any particular proposal. Maybe the right solution is that agents should have a client-side "reader mode" tool, who knows. What seems inevitable is that people will be using LLM-based age
294.
by gwerbin 6mo ago
Interesting. What are your obligations under GDPR in that case? It's not like a packing machine can request data deletion.
295.
by gwerbin 6mo ago
What do you mean by "industrial" in this case?
296.
by gwerbin 6mo ago
You can't control it any more than you can control a draw from a deck of cards, but you can absolutely control the deck of cards that you choose to draw from.
297.
by gwerbin 6mo ago
Citation needed. Sometimes? Sure. Almost always? Questionable assertion.
298.
by gwerbin 6mo ago
Like it or not I think "agents browsing the web" is the inevitable near-term future. Some agents will be malicious, many will not. In 2036, HN posters will be complaining about how such-and-such site only works with closed propri
299.
by gwerbin 6mo ago
> What’s the F is going on? Is the world gone mad or something? Yes, it's madness but it doesn't matter that it's mad because you can't stop it. It's a technological gold rush, with all of the mixed connotations
300.
by gwerbin 6mo ago
As usual, labor saving is only a good idea if the wealth created is distributed throughout society, not redirected to a small group of people.

More ›