esjeon

I agree. Your blog represents you, so it must be fully curated by you. Others' opinions, you can always pull them manually into your website whenever necessary in whatever method and format you prefer. There's no reason to be bothered by random strangers injecting random things onto your own persona. They can always go to other communities if they seriously want to talk about your posts.

Yeah the title was the worst and the most disgusting part, and I'm saying this in totally positive sense. As soon as you catch that "DeepMind" name there, you know it's really about science and a crafted model, not a wannabe general intelligence.

I'm really puzzled by that statement.

RFC 4253(SSH Transport Layer Protocol)[1] says:

   It is expected that in most environments, only 2 round-trips will be needed for full key exchange, server authentication, service request, and acceptance notification of service request.  The worst case is 3 round-trips.

I've never experienced any issues w/ session initialization time. It should be affected by the configuration of both server and client.

[1]: https://datatracker.ietf.org/doc/html/rfc4253

An interesting contradiction is that, in Summer Afternoon, you are in a wide empty town, but, in Messenger, you're stuck in a small planet AND limited vision. You can only see near-by stuffs and never the whole town.

The design of the latter allows packing more details in a smaller space. Also, players are forced to face a lot of unexpectedness, as navigation becomes a natural challenge (which isn't difficult to overcome, which is exactly the point). It's like the blind men and the elephant, and can feel very limited, but you end up using your brain, and, in that process, you fully embrace the world details.

What's being described here lines up with what Jurgen Habermas and Hannah Arendt warned about decades ago. Habermas, for instance, wrote about the "colonization of the lifeworld", where large systems eat into the small, everyday spaces in which people actually build meaning and trust. Arendt, likewise, warned about the fragility of such "space of appearance", where people gather, talk, and act together. The result is "alienation, loneliness, and cynicism or pessimism about the ability to influence future events".

(I really recommend reading Arendt especially regarding how these happen.)

This topic - autonomy - may sound unfamiliar, but it is the essence of democracy and should not be treated as separate from it. They are two sides of the same coin: autonomy is natural small-scale democracy, and democracy is institutionalized large-scale autonomy. While the notion of autonomy is nothing new, revitalizing it in the modern IT era is a bit of an emerging topic. At least that's how I see it.

While I totally agree with the logic, this particular visualization might mislead unsuspecting viewers into believing that our LEO is almost fully saturated. In reality, the satellites are so small that they would miss each other even if placed on the same orbit.

Yeah, this would go no where for quite a long time. But agents will start replacing people once their mistakes cost less than hiring and managing people who perform the same tasks.

This one is where ICE’s operation is completely fair. Koreans have systemically abused ESTA on a massive scale for a long time. Korean companies treat ESTA as the de-facto business visa. US Homeland Security did tighten up the screening last year because of that, but the Korean government chose to do absolute nothing about it. No mention of ESTA & VISA issues was made during the recent US-SK summit meeting. Something like this was bound to happen sooner or later.

And I say all this as a Korean. This article is politically biased rather severely.

Ah, sorry, I did get that part, and my idea goes a little bit further, but somehow I thought I wrote enough.

One thing is that this problem occurs because we have two independent channels that we must independently verify. I’m pretty sure this is a whack-a-mole game, and will never be possible to fix.

Another thing is that, since we don’t trust emails, we hesitate sending links over email. However, the problem here is easy to avoid if services send login links directly to user, and those emails are automatically authenticated by the system.

The actual weak link here is not the procedure itself. It’s the fact that your email services will happily accept phishing mails into your inbox.

I’m pretty sure we can prevent this by issuing some kind of proof of agreement (with sender and recipient info) thru email services. Joining a service becomes submitting a proof to the service, and any attempt to contact the user from the service side must be sealed with the proof. Mix in some signing and HMAC this should be doable. I mean, IF we really want to extend the email standard.

Interestingly enough, suckless folks took the opposite approach with their terminal:

> Goals … Do not reimplement tmux and his comrades.

( From https://st.suckless.org/goals/ )

> avoiding "abuse" just reads as imposing your taste onto all users of your language.

I believe languages are all about bias. A language must represent the preference of the community using it.

We should all learn from the case of Lisp, which is the simplest language and likely the most expressible language. The community suffered from the serious fragmentation driven by the sheer simplicity of the language and tons of NIH syndrome. It took them 30 years to get a standard CL, and another 20 years to get a de facto standard package repository (QuickLisp).

I get the feeling that modern computer systems are so powerful that they can solve almost all well-explored closed problems with a properly tuned model. The problem lies in efficiency, reliability, and cost. Increasing efficiency and reliability would require an exponential increase in cost. QC might solve that cost part, and symbolic reasoning model will significantly boost both efficiency and reliability.

> For Problem 5, models often identified the correct strategies but failed to prove them, which is, ironically, the easier part for an IMO participant. This contrast ... suggests that models could improve significantly in the near future if these relatively minor logical issues are addressed.

Interesting but I'm not sure if this is really due to "minor logical issues". This sounds like a failure due to the lack of the actual understanding (the world model problem). Perhaps the actual answers from AIs might have some hints, but I can't find them.

(EDIT: ooops, found the output on the main page of their website. Didn't expect that.)

> Best-of-n is Important ... the models are surprisingly effective at identifying the relative quality of their own outputs during the best-of-n selection process and are able to look past coherence to check for accuracy.

Yes, it's always easier to be a backseat driver.

I know a publisher who translates books (English to Korean). He works alone these days. Using GPT, he can produce a decent-quality first draft within a day or two. His later steps are also vastly accelerated because GPT reliably catches typos and grammar errors. It doesn't take more than a month to translate and print a book from scratch. Marvelous.

But I still don't like that the same model struggles w/ my projects...

I'm pretty sure they were generally (if not completely) correct when they said that.

It's either the tech is advancing so quickly that many people can't keep up, or simply the cost of adapting outweighs the potential profit from their remaining careers, even when taking the new tech into account.

I've done the same several times with different media. I've used notebooks, wikis, post-its, Obsidian, etc, to organize my thoughts and ideas, but in the end, I've rarely revisited them.

Don't get me wrong - it's still critical to keep track of important information in one way or another. But your own thoughts usually aren't part of that. You are always you, so given the same situation, your future self will likely come up with the same idea you have now (or something even better). That's why keeping track of quick ideas rarely bears fruit.

What you really need to track is unusual information:

- something not from you

- something you can't easily reproduce

- something that sparks new ideas you wouldn't have on your own

In other words, keep the sources of your ideas, not the ideas themselves. This leads to a much lower noise-to-signal ratio because you're more likely to consume well-formulated information, at least much better written than your scattered quick notes.

Note that the username here is a Korean derogatory term for Chinese people.

Possibly, but QEMU is such a critical piece software in our industry. Its application stretches from one end to the other - desktop VM, cloud/remote instance, build server, security sandbox, cross-platform environment, etc. Even a small legal risk can hurt the industry pretty badly.

It's like the problem of half-full vs half-empty. We see LLM can handle certain tasks, but counter-examples are too far from being rare. So, "LLM can do A" is always followed by "LLM fails at A".

The problem is that LLM never performs consistently. It works when it works. It doesn't when It doesn't. No one knows exactly why, and no one can tell when it's gonna fail. For example, even to this day, GPT sometimes gives me wrong calculations, even when it is instructed to use calculator for that. Who knows why it ignores the instruction, nor why it can't reliably perform the addition of two integers. That really screws up with the automation.

Anyways, I'm really tired of skeptic-skeptics. I hate some people believe "half-full" is genuinely better than "half-empty". I refuse that idea completely. It's more about which context you're in. If you need exactly a cup of water, it's half-empty. If you are lucky to have some water, it's half-full. If you have a clear direction you want to go, you discover a lot of deficiencies. If you're there just enjoying the moment, yeah, you can just keep enjoying the new toy.

Anecdotally, ChatGPT still struggles with its own API. It keeps juggling between different versions of its API and hallucinates API parameters, even when I force-feed official documents into the context (to be fair, the documentation is straight awful). Sometimes it totally refuses to change its basic assumptions, so I have to blow up the context just to make it use the up-to-date API correctly.

LLMs are stupid - nothing magic, nothing great. They’re just tools. The problem with the recent LLM craze is that people make too many obviously partially true statements.

AI winters will keep coming as long as the definition of AI stays relative. We used to call chess programs chess "AIs", but hardly anyone says that anymore. We call LLMs "AIs" now, but let's be real: a few decades from now, we'll probably be calling them token predictors, while some shiny new "AIs" are already out there kicking asses.

At the end of the day, "AI" really just means throwing expensive algorithms at problems we've labeled as "subjective" and hoping for the best. More compute, faster communication, bigger storage, and we get to run more of those algorithms. Odds are, the real bottleneck is hardware, not software. Better hardware just lets us take bolder swings at problems, basically wasting even more computing power on nothing.

So yeah, we’ll get yet another AI boom when a new computing paradigm shows up. And that boom will hit yet another AI winter, because it'll smack into the same old bottleneck. And when that winter hits, we'll do what we've always done. Move the goalposts, lower the bar, and start the cycle all over again. Just with new chips this time.

Ah, Jesus. I should quit drinking Turkish coffee.

Yes, but deleting is more difficult and often more expensive than adding. It becomes another burden if you are not bold and determined enough.

> Knowing which problems are worth your energy.

This, but with proper balance. TBH, you can live a happy life if you just stop caring about every technical problem, but that would make you unimaginative and passive. Just make sure your pick a hole (or two) you're gonna die in.

> Run any location experiments outside of California if you want to get an idea of how good your software performs outside of the tech bubble.

I really agree with this because I'm seeing much lower accuracy than what people claim here. I live in Korea, and GPT repeatedly falls back to Seoul almost automatically, and, when I nudge it, jumps to Busan, the second-largest city ~400KM away from Seoul. It's not working so great with other smaller cities and cultural heritages. It fat-fingers a lot if no textual information is present in the photo itself.

GPT also doesn't understand actual geography at all. I managed to get it to nail down which corner of a building is present in the photo, and yet it could never conclude that the photo was taken from a park right across from that corner. Instead, it keeps hopping around popular landmarks in the region, basically miles away from the building it correctly identified. Oh, why, why, why...

Basically it's overhyped rn. It does perform impressively well with clearly visible elements - something anyone can already do with google. It's not like it performs super-human level location tracking. I mean, people can do real crazy things based on shadow details, reflection, items, etc.

I just tossed a group photo w/ some cherry blossom in the background, and GPT immediately answered it's taken in Japan.

Yes, I'm very very very scared. /s

It's more like a lot of people hate CF and that too much of the internet relies on it. It's not like they hate anyone specific.

At the top are some neat high-level stuffs, but, below that, it quickly turns into code-written-in-human-language.

I think it should be possible to extract some more useful usage patterns by poking into related unit tests. How to use should be what matters to most tutorial readers.

The site seems to be down. Likely hugged to death?

Yeah, I totally abandoned Notion last year because it got too slow. The whole thing is pointless if it takes a minute or two just to navigate to the page I want. It takes a few seconds to do that on an archaic corporate SMB share.