Uehreka

5 ms·

31.
by Uehreka 9mo ago
You found it! Yeah the 4th version of the browser is in Haskell and is only a couple hours in, so it’s nowhere near done. The Go version achieved Acid3 compliance in 7 hours, but I expect this one to take a lot longer since Haskell is a bit
32.
by Uehreka 9mo ago
In the past year I’ve used AI coding assistants on a life-saving medical device product (no, followup commenter, I did not ship unreviewed vibes in a medical device product), a tool for editing documentation used in healthcare (no, followup
33.
by Uehreka 9mo ago
There are frequently claims that Anthropic is somehow diluting or dumbing down models in some subtle way. Unfortunately it’s tough to validate these claims without a body of regularly checked evals. This test set should hopefully help settl
34.
by Uehreka 9mo ago
> That will change real quick if everyone loses their job to AI. No, this is exactly my point: they will be angry, unreasonable, and thirsty for revenge. They’ll hand over freedoms like Halloween candy. How about a law where the governme
35.
by Uehreka 9mo ago
> Making up jobs to keep people employed isn't a viable solution to me. Supporting them in some way (re-training, UBI, service work, ...) seems like the only way forward. If this is what you think needs to happen and you live in the
36.
by Uehreka 9mo ago
I was more concerned about the “damning” part. I don’t think there’s a lot of solid reasoning in the article to point to any wrongdoing. If Steve Yegge and Pete Steiberger ignored this I wouldn’t blame them.
37.
by Uehreka 9mo ago
I would not call this “basic automation”. I’m also not saying “this dude cracked the orchestration code” (you’re free to be mad at people who are, but I feel like it’s more interesting to engage with the people who aren’t). I will say that
38.
by Uehreka 9mo ago
Is it? I’ve been experimenting with Gas Town, I’ve learned a lot of new ideas about coding agent orchestration, I haven’t given Steve Yegge a dime or touched any crypto. How am I being scammed? The article just seems like yet another “look
39.
by Uehreka 9mo ago
> (e.g. implement a well documented wire protocol with several reference implementations on a microcontroller) I have had an AI assistant reverse engineer a complex TCP protocol (3-simultaneous connections each with a different purpose,
40.
by Uehreka 9mo ago
I (and I suspect many others) usually think of a constitution as “the hard-to-edit meta-rules that govern the normal rules”. The idea that the stuff in this document can sort of “override” the system prompt and constrain the things that Cla
41.
by Uehreka 9mo ago
By the time I switched to GPT 5 we were already on 5.1, so I can't speak to 5.0. All I can say is that if the answer came down to something like "push the bias in the other direction and hope we land in the right spot"... wel
42.
by Uehreka 9mo ago
GP here: My problem with a lot of studies and data is that they seem to measure how good LLMs are at a particular task, but often don't account for "how good the LLM is to work with". The latter feels extremely difficult to q
43.
by Uehreka 9mo ago
It feels like a lot of people keep falling into the trap of thinking we’ve hit a plateau, and that they can shift from “aggressively explore and learn the thing” mode to “teach people solid facts” mode. A week ago Scott Hanselman went on th
44.
by Uehreka 9mo ago
“Disallow writes” isn’t a thing unless you whitelist (not blacklist) what your agent can read (GET requests can be used to write by encoding arbitrary data in URL paths and querystrings). The problem is, once you “injection-proof” your agen
45.
by Uehreka 9mo ago
> I figured it's how they were able to weaken a chess engine back in the day; can't adjust the overall strength, so add random blunders. In tom7’s Elo World, he does this (“dilutes” strong Chess AIs with a certain percentage of
46.
by Uehreka 10mo ago
I normally wouldn’t be this pedantic, but given that this is a conversation about pedantry it only seems right: you’re using i.e. and e.g. backwards.
47.
by Uehreka 10mo ago
This clearly was not what GP was talking about. Believe it or not, not all people do things for purely cynical reasons.
48.
by Uehreka 10mo ago
It is not in the interest of “people who complain about things being too far left” to get specific. To do so can only increase the number of people who realize they disagree with them. The vagueness is purposeful.
49.
by Uehreka 10mo ago
The definition of “Tech bros” is “tech people you don’t like”. There’s no agreed upon definition (just like how people disagree about what is/isn’t a “grift”) because it’s not meant to be descriptive, it’s a rhetorical device.
50.
by Uehreka 10mo ago
I’ve gotten so frustrated with AE the past couple years that my New Year’s resolution is to bite the bullet and try Resolve for my next couple projects. I’m even willing to splurge for the license if it means I can stop paying Adobe every m
51.
by Uehreka 10mo ago
I want this to happen, but this past month I tried using Blender’s Video Editor (after hearing the great news about Compositor Modifiers) and boy oh boy does it chug. Apparently Compositor Modifiers only run on the CPU, so trying to apply t
52.
by Uehreka 10mo ago
> (as a hobbyist who likes cooking up some vis for my DJ sets) I generally get really annoyed when I hear someone say that a particular piece of open-source or free-as-in-beer software “blows After Effects out of the water”[0], but not h
53.
by Uehreka 10mo ago
Don’t most camera manufacturers (like ARRI and BlackMagic) have test footage for their raw and/or log formats on their websites? Here’s ARRI’s (which includes ProRes in addition to their proprietary formats) https://www.arri
54.
by Uehreka 10mo ago
From what I’ve read, you want to make sure that the setting is spelled FILMMAKER MODE (in all caps) with a (TM) symbol, since that means that the body who popularized the setting has approved whatever the manufacturer does when you turn tha
55.
by Uehreka 10mo ago
I was doing something at work that involved calculating round trip times from/to Android devices, and learned that although it should be possible for NTP to sync clocks with below-second precision, in practice many of the Android devic
56.
by Uehreka 10mo ago
> It would be hard to think of another programmer whose body of public work is more impressive than FB. Not many, but these do come to mind: Linus Torvalds, Ken Thompson, Dennis Ritchie, Donald Knuth, Rob Pike. But yeah, it’s rarefied ai
57.
by Uehreka 10mo ago
I’m not GP, but I’d argue that “making frontier AI models more offensive in black hat capabilities” is a thing that’s going to happen whether we want it or not, since we don’t control who can train a model. So the more productive way to rea
58.
by Uehreka 10mo ago
Nah, that doesn’t work when the substance of the change is this intense and has an actual effect on peoples’ bottom lines. If they wait a few months and try again, people will see their bills go up immediately and they’ll all get mad again.
59.
by Uehreka 10mo ago
I would expect open weights models to always lag behind; training is resource-intensive and it’s much easier to finance if you can make money directly from the result. So in a year we may have a ~700B open weights model that competes with G
60.
by Uehreka 10mo ago
No, it’s an example that shows that LLMs still use a tokenizer, which is not an impediment for almost any task (even many where you would expect it to be, like searching a codebase for variants of a variable name in different cases).

More ›