Centigonal

Worth noting this was GPT-4o, which is perhaps the most syncophantic and suggestible major model released by a frontier lab. It was the subject of this entertaining and insightful Eddy Burback video: https://www.youtube.com/watch?v=VRjgNgJms3Q

The 15 people charged in the case this story is about were not overstaying their visa; they were protesting ICE after ICE killed a US citizen in Minnesota.

Not a Googler, but from what I've heard from colleagues at GCP, it's is a very different organization than much of the rest of Google.

For this reason, I think it's important to listen to testimonies from both: people who've succeeded at the thing you want to do, as well as people who failed. That way, you can get a grasp of the common traits of successful examples while also understanding the most common failure modes.

This is some next level legal judo. Say what you want about the metaverse or muse, but it's clear that Meta is still innovating when it comes to its Terms of Service.

Clearly, some readers did, given all of the similar posts on the HN thread.

It's also worth noting that the article has been edited in the last hour. Some section headers read a lot more clearly and naturally now (which is awesome!).

I think it's possible to give someone their flowers for being a great innovator and also call them out for producing writing that's painful to read.

Appropriate username!

> If you don't want to be "locked in", you'll likely have to pay for that in some other way.

That "some other way" is almost never via the wallet. From Apple to Whoop to Snowflake to Oracle, the most expensive ecosystems are often the ones with the most lock-in. More often than not, the "some other way" you pay is by having to put up with janky UX and incomplete featuresets.

I saw a similar notice on a random wall at the office of a late-stage startup I occasionally visited (since acquired) -- although it was in regular sized font. I actually informed one of my friends who was looking for a job to sponsor his visa about the position, and he was hired.

I think you're confusing corporate fealty with the opportunity to make an impact through innovation. I think people like Shuji Nakamura and Carl Bosch might have said similar things about their jobs.

Despite being superficially very similar to Anthropic, OpenAI seems to consistently generate more drama. I'm not sure what this means, but I think it's significant.

This is quite good, especially given the size and the fact that it runs on the CPU.

Accepting Vance's premise, is he saying that the solution to replacing American workers with indentured servants is... removing the path out of indentured servitude? It sounds like they're not touching H1-B itself, just the path from H1-B to residency.

TECH! PRIESTS!

https://warhammer40k.fandom.com/wiki/Tech-Priest

out of core sounds awesome! biggest thing that forced me to switch from pandas/polars to other solutions back in the day.

new entrant in this weight class, US lab.

Do you have any idea what Muse created that set off the moderation?

OHHH - yeah, that's messed up. I didn't realize it did that!

Oh, I didn't realize it's broken! I thought you were just surprised that the header stayed in place while the table scrolled.

I envy your lack of familiarity with excel/sheets

haiku (the model family) was not inherently inferior to its larger, more expensive siblings. Sometimes you prefer a cheap, fast model, just like how sometimes you prefer a shorter poem.

This feels like the steam catapault kerfuffle from earlier this year. Maybe I'm missing some vital reason these retired technologies must be revived, but it's not inspiring when our military doctrine is "let's go back to what worked 60 years ago"

wouldn't less compute result in slower inference, rather than worse performance?

I don't know what Sun's strategy was in the 2000s, but doing all of the R&D to develop exotic new hardware and then focusing the sales force on software seems like they're getting the worst of both worlds.

I think the point he's actually making is that if we start training AIs or designing AIs as if they're conscious, we might be more likely to create future AIs that are either very good at persuading people that they are conscious or AIs that are actually conscious, which will have major societal repercussions.

Update, I am less convinced of my earlier comment in light of the Gemini hack: https://www.wsj.com/tech/ai/gemini-hacked-three-companies-in...

Yes, but then we collapse two different groups with isolated influences and philosophies into one group, with the associated penchant for groupthink.

I'm beginning to believe that, in our hyperconnected world, the best way for a group of people to explore a wide surface of ideas might actually be to deliberately scale back inter-group communication and intentionally subdivide into silos to avoid all converging on the same thing.

Some hot water radiator systems use wax actuated valves made from brass (Taco brand is common). They are extremely simple and robust.

This is about chats, not calls though - you can look at the context and make the same determination.

Even for calls, "hey, do you have a sec to help me troubleshoot the CDN setup?" is way better than "hello," followed by a call if you respond.

How are you determining how much oomph Claude gives a particular question? When I ask Claude Chat to e.g. summarize a document, it does this and returns a summary inline, whereas cowork does a bunch of tool calls and intermediate steps, eventually writing a document to disk. How does new claude know how deep to go?