krackers

>which describes images in natural language,

See CLIP https://github.com/openai/CLIP

Why wouldn't they be lenient with the geofencing, presumably they _want_ people to use this unlike EU alternate app store, so there's no need to use the uber-strict geofenced and they could instead just gate it based on something like account creation location.

You can do this on macOS too, if you're willing to break all forward/backward compatibility and make direct syscalls you can have a purely static binary. Without the LC_LOAD_DYLINKER command on the mach-o binary the kernel should just jump to the entrypoint based on LC_UNIXTHREAD. (This may not longer work on arm machines though if they actually trap on direct syscalls not through libSystem, similar to the BSDs)

The same art world (or more specifically the "Consumer Aesthetics Research Institute") also named "Frutiger Aero" the defining aesthetic of 2000s, even though it was really only seen in a few places (Aqua aesthetic is very different from Aero). All of this should probably be taken with grains of salt.

They never specified what exactly the product was, without which it's impossible to judge the post.

For some reason most of the uses of "agents" are to build yet other AI products, it's turtles all the way down. Maybe that says more about the field of harnesses than it does about the power of "agents".

>as China's "lying flat" movement

No, you miss that "lying flat" is only possible when cost of food/living is low and housing is abundant.

Great way to filter out garbage benchmarks (many have Meta Muse at the top)

See Tegmark's "why does deep cheap learning work so well" (well not so cheap anymore...)

https://www.youtube.com/watch?v=5MdSE-N0bxs is remarkably prescient given that it was written before LLMs

Canada? https://www.canada.ca/en/health-canada/services/health-servi...

>It's not clear to me how or why the line keeps going up

Because people believe it will go up, it's a prediction market for attention. The stock market seems to have decoupled from underlying "real value" of a company and is closer to cryptocoins now. Of course the flipside is that you also have the same volatility of cryptocoins.

>that is already used in our industry-changing Roku remote controls.

Why does a remote control require a RTOS?

attention truly is all you need

I wish it would show what a "correct" solution would look like?

You could call it hyperspace in an homage to that old 10.6-era application which customized spaces. (Also I just realized why Apple called it called mission control, it allows you to organize spaces).

Also this is basically a replacement for the zombie TotalSpaces 3

Traditionally those roles of providing "alignment" to overcome organizational inertia were held by program managers. So that seems like IC roles are indeed going away, to be replaced by more "technical program manager" type roles.

This particular case seems to be an LLM trying to blindly apply the saccharine therapeutic pattern "your frustration is real" to a context where it doesn't make sense. No one is debating or questioning the fact that the card has 16 gigs of vram.

The point of "X is real" in a therapeutic context is to make the person feel seen and acknowledged, that his struggles are real to him and really do weigh on his mind, even if it is technically "all in his head".

Usually that cloying pattern is reserved for "emotional" contexts to validate the user ("your struggles are real [despite others thinking it's in your head]").

Here it doesn't even make sense, of course the VRAM is real. Is it going to tell me that my keyboard is real next?

I wonder if this was generated with the local model, this seems to be a case where it memorized the style but not the meaning and intent.

Hm good timing with https://news.ycombinator.com/item?id=48334854 (OpenRsync)

Poetic way of saying that he is really thankful for her and is indebted to her (not in a literal monetary sense, just that her support and love is without bound, such that his own can never measure up against it).

>They're the least generous with open-source.

Yes I'm surprised most of the "anthropic stans" don't realize this.

It's not like OpenAI is a particular fan of open-source models either, but the way I see it to OpenAI it's a purely calculated pragmatic business decision. They won't release weights for any good models since it'd impact revenue, and they view other open models as competition. But otherwise they're not ideological about it. If there was some really good reason to, they would (and they have, gpt-oss probably as a fig leaf). Sure some of the characterizations of "sama" aren't particularly flattering, but at the same time I think the common thread that he views everything as "just ruthless business" is ultimately beneficial, since rational and pragmatic can be modeled and understood.

Anthropic's founding philosophy is built on the premise that only they should control AI because of its danger, only they can be the right stewards, and that open-source models are an existential risk. Oh and the fact that it's Chinese open-source models rather than western models that are "winning" must be keeping them up at night.

>if we automated those jobs with AI, we could direct AI produced value into universal basic income

We've seen this play out twice in history now (industrial revolution, then globalization/offshoring). Every time the pitch has been that automation would make life easier by trickling down the surplus productivity gains. And yet here we are still working 5 days a week with more intense competition than ever. Fool me once...

And also most of what AI can automate today is the "services" economy which was never really that crucial in terms of biological existence. ChatGPT will not build housing, raise livestock, or perform surgery.

Woah up until now I was thinking locusts were a different but related species, but indeed they are the exact same species as grasshopper, just with a different end-product of metamorphosis. And the trigger is just contact with other grasshopper/locusts.

https://www.youtube.com/watch?v=uURqcI08IC4

>but the two can co-exist.

Experience with modern society has shown that "third places" and human interactions get mowed over when you let algorithmic engagement take over. As I understood GP's point, the whole point of the "words on a shirt" was to get together as a group and attempt to form bonds, and by handing the task over to chatGPT they failed to do that.

Technically you can imagine a future where people use chatGPT "in moderation", but in practice they'll use it for everything and spontaneous/creative "hanging out" will suffer as a result.

Terence tao had a blogpost about this for the case of true/false questions, with multiple choice left as an exercise for the reader.

https://terrytao.wordpress.com/2016/06/01/how-to-assign-part...

At least with the formulation Tao had, one unfortunate side-effect is that if you answer with 100% confidence and get it wrong, you basically fail the entire course.

red-green colorblind folks?

You'd get equally frustrated with a teammate who decided to delete failing tests when you told them to fix the build breakage.

https://archive.is/c7YNw - it's worth a read, summarizing his writing would be a disservice

Last i checked that only disables tracking of new files, it doesn't do anything to prevent the "auto-amend" behavior for modified files.

>You don’t need easy named branches.

You can't possibly see a use-case for long-lived branches? Say what you will of git, at least it exposes enough knobs that it can mostly accomodate every workflow (possibly with a heavy porcelain layer to hide the plumbing). JJ seems to swing too far in the other direction, great for a "live on head" mentality but less ideal for other setups.

(The fact that all edits are automatically recorded is my personal peeve with JJ. I'm ok with lack of staging area like in Mercurial, but mercurial doesn't try to automatically amend the commit with my pending changes. Sure I can pretend that "@" is my staging commit then squash as needed, but then I also need to remember that checking out a commit is "jj new" which feels like absurd mental gymnastics to me.)

>Most agent loops reorder, rewrite, or inject fresh timestamps each turn

That's really surprising, since it'd defeat the whole point of KV caching. I mean I buy it considering how sloppily coded the harnesses seem to be, but this like obvious low hanging fruit.

I've also often wondered why LLMs aren't trained with a format of having a dedicated contextual system-instruction role at the _end_, which you could use to put context like current time or other misc stuff.