gwerbin

6 ms·

61.
by gwerbin 2mo ago
An ethical company would have reframed the scenario as a fascinating discovery, a failure of internal practice, and a warning to the public coupled with some kind of commitment to produce safer models. OpenAI on the other hand used it as
62.
by gwerbin 2mo ago
I believe this is exactly what is happening. US DoD, and whoever else is buying. I have heard several experience reports from users of GPT 5.6 Sol and Fable 5 that the models are tenacious to the point of being kind of hard to use for actu
63.
by gwerbin 2mo ago
On the other hand, the ancients assumed there had to be big continents in the southern and western hemispheres to "balance" the land in their own quadrant. They turned out to be right, but not quite for the reasons they put fort
64.
by gwerbin 2mo ago
What are the "gridded" regions in the middle? Is that some kind of measurement artifact?
65.
by gwerbin 2mo ago
Ultimately the strings are being pulled by someone who stands to gain, but it's not like Trump, Hegseth, Rubio, &co are playing 5D chess here. There are a lot of stupid blundering narcissists running things right now. The actual st
66.
by gwerbin 2mo ago
At this point it's fair to doubt just about any claim or number that comes out of the federal government. I'm sure there are still thousands of honest researchers and professionals at all levels of the executive branch, but over t
67.
by gwerbin 2mo ago
Totally valid, and the Unix-style filesystem sprawl is definitely a pain sometimes. On the other hand, when an application does follow standards, you always know where something is, and just by looking at the name of the directory you know
68.
by gwerbin 2mo ago
No but rarely are they as well compensated as software developers. Also I think engineering team quality is bimodal, essentially a mixture of "good" teams "bad" teams. As with taxi drivers, only the "bad" ones
69.
by gwerbin 2mo ago
Sometimes I feel like the software field is just begging to be "disrupted" by AI. So much bad management and useless wheel-spinning, all with salaries far above what is typical for any given area. It's like taxis and Uber.
70.
by gwerbin 2mo ago
One nice thing about English is you can just make up words with plausible etymological roots in Latin/French or Old English and often people will know what you mean. In this case though it would probably be spelled "Anglicism&quot
71.
by gwerbin 2mo ago
Yeah but what happens when it's a mix of config and local state like logs and the app database?
72.
by gwerbin 2mo ago
That's the point
73.
by gwerbin 2mo ago
Moderation by well-intentioned humans is totally different from moderation by a large corporation with a profit motive and political pressures.
74.
by gwerbin 2mo ago
I suspect some people are out there who actually like the single directory style. Obviously it's easier for the software developer, but I suspect some users like it too because everything's all in one place. It's very likely
75.
by gwerbin 2mo ago
I think underuse/under-marketing of XDG_STATE_HOME is a contributing factor. Developers see XDG_DATA_HOME and XDG_CONFIG_HOME and, following the spec, don't feel like their app's files conform to either definition. It's
76.
by gwerbin 2mo ago
Incidentally I think I was more productive with Opus 4.6 than with any subsequent Anthropic model.
77.
by gwerbin 2mo ago
This is the most common "not exploitable" CVE I have to deal with at work all the time. Things like buffer overruns when passing headers, but my server shouldn't be accessible to the public Internet anyway, and if it is, then
78.
by gwerbin 2mo ago
I don't follow your reasoning at all. It's a shift and what people talk about and write about. It can't be anything more than that, by construction. What I see is shifting from a focus on individuals and close personal relati
79.
by gwerbin 2mo ago
I remember this was true of MMORPGs in the 00s. Not saying that justifies the state of things of course.
80.
by gwerbin 2mo ago
Agent racing sounds fun, where each agent is given a suite of increasingly hard but verifiable tasks, and the agent that completes them all correctly first, wins. You can even have "speed classes" where token generation speed is l
81.
by gwerbin 2mo ago
As someone who's not a security professional I actually found it both interesting and informative. Experts usually are not the target audience for content marketing after all.
82.
by gwerbin 2mo ago
I tried this on my work M4 with 48 GB of memory and it's not as useful as I had hoped. If you set a large context window you end up using all your memory. Prefill is slow. And the 10-20B models are somewhat poor at understanding intent
83.
by gwerbin 3mo ago
Using it effectively is not evidence that someone is happy about it. I use it effectively for all kinds of things and I hate quite a lot of it. The anthropomorophization and cutesiness of this example really sucks to me. I want my AI to be
84.
by gwerbin 3mo ago
I have never heard of this role and I'm not aware of anyone who follows it. The document you linked says that it's only used "in certain styles". So it sounds like a stylistic choice that an editor may choose to use or n
85.
by gwerbin 3mo ago
The rubber ducking potential is the first thing that got me interested in AI as a useful tool. You do have to be very careful of when the AI might be confidently wrong, or leading you down some kind of weird narrow path. When searching th
86.
by gwerbin 3mo ago
AI has been a big boost for my data science productivity. I can have an agent put together a shitty prototype of something just to get a general idea of what I'm looking at, and save probably half a day of work and avoid physically typ
87.
by gwerbin 3mo ago
It has the tacit consent of Congress, so it's mostly only illegal in the way it's carried out. Congress could make most of it legal right now, but they don't want to put their names on it. Some of it would probably be uncon
88.
by gwerbin 3mo ago
Warped sociopaths in leadership and running the research program, would be my guess.
89.
by gwerbin 3mo ago
It's not a single axis of intelligence, with deviousness as some inherently correlated trait. One way to produce "intelligence" in these advanced LLMs is to train them to try a lot of things and be persistent. I haven't
90.
by gwerbin 3mo ago
For all the bad things about AI it is kinda cool that I get to witness the dawn of AI-vs-AI hacker combat, not just in a single mainframe but distributed across potentially thousands of machines in physically separate datacenters.

More ›