kmacdough

Its not about geometry, it's a parallel compute thing thing. CPUs typically don't have more than 10 or 20 cores. GPU have 100s to 1000s.

Matmul is very well parallelized. More, lower power cores will always pay off handsomely.

The problem is simple: qualified immunity has become a blank check. The officer can simply claim they didn't know the law. They somehow can't be expected to understand basic constitutional protections.

The government is not a monolith. Being owned by the city doesn't have to mean the cops are in control. The municipality can determin by law exactly who operates the infrastructure, who has access to what, what process they must follow, and how that all will be monitored and enforced. "The government didn't handle this well, therefore they can't be trusted for anything like it again" is a misunderstanding of how governments are constructed and how power can be separated between legislatively mandated structures. Find the source of the abuse, then build a structure to check that abuse.

Sold as learning, but that was a marketing term, not a technical one. From a technical perspective, the LLM is not learning. Only reacting based on its original training.

You might argue that the systems we've built around them are learning in a way, as they strategically condense and save artifacts from past interactions to pass into the LLMs context. But the LLM itself, which is the source of the intelligence, is not learning. It remains entirely unchanged throughout inference. This difference may seem trite, but it has significant impacts over the long term behavior.

Context is not the same as learning. It's easy to conflate because they're tightly coupled in our brains.

The underlying structure and tuning of the LLM are entirely unchanged by context. It merely affects the attention and activation of the network. The LLM will not be able to work with this hypothetical new language unless it is in context. This does not fit the computational meaning of learning.

Smart is not a well defined term. Nor is it's general idea formally understood. Use it freely, but you won't be saying anything meaningful unless you define your usage.

Unfortunately doesn't cover scrolling HN while the agent toils away.

With a striking lack of numbers, I'm not confident. I my experience, everything underspecified in a marketing release is unflattering. They're also not a chip designing company, but they're probably trying to keep up on the eyes of investors. As the article mentions, several of their competitors are chip designers and already have working procuction inference chips.

Right, but the entire damn point is privacy protections enable people to be more honest. The entire point is supposed to be good data so we can make informed population wide decisions.

And race is a pretty big one under the current administration which has had hundreds of legal immigrants arrested for weeks to months off of "suspicion" that for lack of concrete evidence could only amount to racial profiling.

I believe there is hard evidence that role-playing prompts are effective at leading it towards particular strategies and trains of thought. Not sure that SWE has been specifically studied, but proper science is very slow in the context of rapid change and broad context. It's good to stay grounded in the science that has been done, but we're going to have to do our best in uncharted territory for a while.

"Don't make mistakes" does seem dumb. It's not guidance.

Yeah just watch out, they're trying to eat your 401k and they've got a powerful easily influenced friend.

Hate to break it to ya, but this is how most C-suites operate. Their job isn't to run a company well. It's to appear to the board/investors that they're running a good company.

It is a better play to do the popular thing in a way that measures as "ahead". Then it's hard to argue against a raise. But if you stick your neck out on your thoughtful expertise, it can take years or more for the value to come thru. You can easily be replaced by then.

The only antidote is a board that has a real working nuanced understanding of the entire industry. But this rarely happens, for many reasons.

This is a separate issue, I think even Bambu Studio can't connect to printers in LAN mode on a different subnet.

Yes, that's the point. The nerworking is broken. The issue isn't unique to a specific slicer, their software sucks. Orca ran into the issue because they wanted to make a basic feature that works on every other printer on the market work on a bambu.

This formatting is intolerable on mobile.

I learned much of what I know about computer and low-level systems engineering from Minecraft. Watched lots of videos making CPUs and built many components myself including a full ALU with a look-ahead adder and hardware multiplication.

It's still circular. They will succeed or collapse together. And since they make up such a fraction of global market cap, we're in for the ride together.

And the circularity makes the actual investment numbers fairly meaningless. They don't mind if they end up overpaying for future services, as long as they overpay each other equally.

They have to. The device storage is itself encrypted, so the FBI already broke into the phone. When the device is unlocked, notifications are visible by design and therefore available in plain text to the user. The edge case is with disappearing messages, a feature Apple did not build for. The message is intended to be plainly visible to the user, but only for a controlled time on the assumption that the users privileges may eventually be compromised.

This makes for a very odd and specific interaction with a 3rd party feature. Security is a hard problem.

I'm confused. What's your point?

Obviously economies that rely largely on second hand technology are going to have old technology. Much of Africa is in this bucket. But looking past the extremes, India is at nearly 80% right alongside Germany. They fall in very different average income brackets. So the correlation isn't tight.

I can't see any value in pointing out vague correlation between income and proliferation of a new technology. It's the most obvious of observations.

Given adherence is a more significant practical barrier, it's probably the better signal. That is, if we decide too look for signal here.

Criticizing the government is not hostility. Its wanting to move towards a better country. This is EXACTLY what the 1st amendment is intended to protect. Whether the legal system decides it applies here is one question, but there are heaps of documents and communications between founding fathers and other figures making this clear. Many of those folks were immigrants themselves. So the idea that it wouldn't apply to legal immigrants is wildly out of line with the founding ethos of the country.

Everything in context. This is one of many reasons I'm a proponent of squash-and-merge. If a change really needs more than one permanent commit, it should probably be split up or if absolutely necessary should be on a feature branch maintaining similar process. Under this process, feature branches are not squashed.

This leaves developers to commit locally and comment as much or little as they like.

Two things can be true. And often that's precisely when we lose out with modern engineering that is much more single-minded.

Their only advantage...

How are you coming to this conclusion?! Their warmer has very meaningful effects on processing, attention and other visual effects as is the point of the discussion in the first place. It's not clear what makes you so sure that color differentiation is essential and the other effects are irrelevant.

No I absolutely don't know what matters. But it seems neither do you.

But the discussion is about the safety of systems being actively implemented. For now, people are required to maintain data centers. Right now, these centers are transitioning to high voltage DC. Right now the best attempts at fully automated data centers are in progress with unverified status. So, for now we need to keep the humans safe. Future data centers not needing humans doesn't keep the current humans safe. Thus, optimistic future speculation is off topic and not relevant to the discussion of human safety with high voltages.

People are tired of off topic speculation masquerading as relevant to real world problem solving. It's not a hate boner.

EDIT: expanded to make it clearer why I believe the speculation to be wildly off topic for this particular thread.

The concern isn't keeping people out of the facility altogether. It's keeping people safe when they are hot swapping dead or outdated rack units.

You could consider a robotized approach, but for now robots aren't very good at this type of task. Whether the robotized system is in space is irrelevant. It's not a "next five years" type of solution.

I'm all for this. Unobtrusive and just enough to put it in folks mind. I'm not a big fan of the Author though.

How come there's no photos of the before/after? That would be much more useful than the condescension. They treat the reader like they've already whined with the least reasonable complaint. But they don't even bother to make their case properly, just point fingers at who else has done it, as if that's reason enough.

Wikipedia is NOT a good comparison. Their banners are obtrusive, obnoxious and the reason I stopped my $1/mo donation more than a decade ago. Well more specifically the begging/crying prompted me to look into their finances and spending and it turns out they were pretty irresponsible at least then. Lots of highly paid "administrators," software devs working on poorly thought out systems. Lots of interfering with volunteer contributions.

Well that is a high enough level even the bet would probably be enforced. But the point is valid, this is a slippery slope and there's a reason these bets are illegal.

What are we testing here?

It feels like a very odd test because it's such an unreasonable way to answer this with an LLM. Nothing about the task requires more than a very localized understanding. It's not like a codebase or corporate documentation, where there's a lot of interconectedness and context that's important. It also doesn't seem to poke at the gap between human and AI intelligence.

Why are people excited? What am I missing?

Based on who's definition? Plenty of people use "middle management" to describe anyone under executives, but who doesn't spend time closely aligned with front-line work. In some structures, lowest level management is closely integrated with front-line, but I'd argue PHB has next to zero exposure to the actual work or goings-on.

The biggest advantage? Regional caches.

If your traffic volume stays well below server capacity, the network connection is impeccable and throughput isn't a major factor in ISP costs, then the improvement is indeed negligible.

However, if the server or it's network ever experience congestion or disruption, or you simply face sheer volume, that's when CDNs shine.

Regional caches insulate clients from network or performance issues with your server or its connection. If you have high volume, they also drastically reduce traffic through the backbone which can reduce ISP costs, depending on your contract.

Like others say, this really applies to all static content, including static media, css and js.

I hear you ask, "how can you be exceedingly good at this?" Caching itself isn't tricky, but catching efficiently and effectively at such huge scales is. The fact that Cloudflare offers these services for FREE, is a pretty good (perhaps Faustian) deal.

I'm assuming this is referencing "taking a pound of flesh" generally meaning to being cruel in demanding what you're owed (from Shakespeare Merchant of Venice). Presumably they're tired of unloading on people for not following thru or contributing. Doesn't seem like the best use here, particularly so indirectly.

Furthermore, paid models are heavily subsidized by bullish investors playing for monopoly. So that tips the scales further towards Deepseek.