variadix

Even worse, people will discuss/diagnose/troubleshoot with their LLM of choice, some of which will be valuable, and the only way to get access to said information will be through the LLM provider’s paywall. I think we are entering an information dark age, for many reasons, that will likely only get worse from here on.

Perhaps, but I think what people mean by intelligence is something that learns and adapts. If LLMs couldn’t do in context learning I don’t think people would think of them as AI, more as a kind of queryable database via natural language. There are other algorithms that learn and adapt, but in much more narrow circumstances that most people won’t obviously interact with.

Yeah I don’t understand this port at all other than as a big marketing stunt.

It would have made far more sense, for reliability, efficiency, cost, etc., every metric really, to use or write a source to source translator that preserves as much structure from the original code as possible. Typically if you do a rewrite there are lessons learned from the existing code base that you want to take into account when doing the rewrite; using a bunch of agents to do the porting file-by-file buys you none of that. In either case the code will be an unidiomatic translation, just with LLMs you get an added source of indeterminism and a huge bill at the end of the month.

Alienware does the same thing, and as far as I could figure out, there is no way to stop it. The enshitification of Windows at this point is incredible.

Agreed on Sedra and Smith vs Razavi. Razavi uploaded lectures to YouTube that helped me a ton at the time.

One interesting feature of signed vs unsigned in languages where signed overflow is undefined is that signed numbers _behave_ like mathematical integers, whereas unsigned integers (because overflow is valid behavior) _do not_ behave like mathematical integers. For example: if you write a loop that sums numbers from 1 to N, for signed integers the compiler can assume the expression won’t overflow and emit the standard mathematical closed form, because unsigned can overflow it has to account for this and use a different closed form (if one exists for the unsigned case). IMO unsigned should only be used when you want to manipulate the bits of an integer and not the integer itself, but C and C++ fight you when you try to do this because of how ingrained size_t is.

I personally use a 3x6 Corne for non-gaming tasks. I would recommend it for anyone that uses a keyboard for more than a few hours a day to preemptively reduce the risk of RSI, especially programmers. The small layout and tenting reduce wrist and finger strain significantly, to the point where when I have to use a regular keyboard at work I can quickly notice the pain in my right pinky. If you want something with curved key wells the Dactyl Cygnus is a Corne-like option. I haven’t bothered to try anything smaller since I don’t see the point (for non-steno), as all keys on a 3x6 are within 1u of the home row.

Indeed. Their single most important piece of evidence not providing any strong indication the decline is from LLMs does not bode well for their argument. If it were the case you would expect the slope to become more negative with increasing LLM capability, and it in fact does the opposite.

I largely agree. Split, ortho/columnar, and tenting is 90% of the strain benefits. The biggest benefit from layout change (for programmers at least) is moving symbols and numbers to a layer where they share positions with alpha keys so you don’t have to extend to type them. Any benefit I could imagine from switching alpha layouts (more even utilization between left and right hand for typing prose) is mitigated by the switching cost when trying to use a normal computer.

There are momentum indices. Momentum is actually a strange phenomenon from the perspective of the efficient market hypothesis because it does not have an obvious risk to balance its premium compared to other empirical factors.

Immigration is not a solution to low fertility. It might be a solution to labor being expensive, particularly if you’re a capitalist, but it doesn’t solve the problem of low fertility among a native population.

They can restrict it to US citizens only: just do KYC and enforce it. Any citizen exporting ITAR capabilities would be committing a felony.

I think it is inevitable that more capable models will require export controls, KYC, background checks, verification of credentials, etc. Even if you don’t buy Anthropic’s marketing of their current models, there is a future where models are capable of developing chemical weapons, biological weapons, cyber weapons, etc. which are genuine security threats governments will care about controlling and preventing. Maybe this prevents big commercial labs from developing more capable models until they figure out appropriate safeguards, but this is a good thing, a world where you can ask an LLM for an airborne rabies genome and it will simply produce it for you is not a world we should want to live in.

You can probably eliminate the switch bounds check by making the default case execute __builtin_unreachable(). The switch version is safe for non-conforming opcodes where the computed goto version invokes UB.

People have built entire careers shipping garbage, now they can ship 10x more garbage and to them that’s progress.

The STL is not good if you want performance or predictable behavior. The issue is in the specification and the requirements placed on certain algorithms and data structures. It’s easy to beat unordered_map for example with an open addressing hash map, small vector optimization can’t be implemented in vector due to standard requirements, etc.

The consumer economy only exists to extract value from common people and funnel it up the wealth ladder. If robots and AI take over all the production, you don’t need a consumer economy, the robots produce and their output directly goes to the top. The rest of us are left to starve.

This is a useful lesson! Tuning an allocator for a single application reduces to optimizing for that application’s empirical allocation patterns (sizes, lifetimes, access, usage).

I’m not exactly sure what you’re getting at, but I think I’ve had a similar question: why don’t HDLs have language elements more representative of what digital circuits are constructed from, namely synchronous and asynchronous circuits, rather than imperative input triggered blocks (processes IIRC, it’s been a while)?

I always thought it was confusing to design a circuit mentally (or on paper) out of things like muxes, encoders, flip flops, etc. and not have language-level elements to represent these things (without defining your own components obviously).

I remember looking this up, and I believe it’s because the languages were originally designed for simulation and verification, and there are things you might want to do in a simulation/verification language for testing that are outside of what the hardware can do. Mixing the two is confusing IMO, but clearly demarcating the hardware-realizable subset of the language would be better than the current state.

Maybe, another possibility is the frontier providers change their pricing terms to try to capture more of the value once a sufficient number of people’s skills have atrophied. For example: 20% of the revenue of all products built with $AI_SERVICE. For someone several years out of practice they may have no other option.

Insiders bring information to a market. Intelligent analysis and prediction also does, but obviously insiders have special information they are incentivized to bring to the market. Most people placing these bets are simply gambling, insiders and analysts at least have rational reasons for placing bets and add information to the market.

This is a fundamental misunderstanding of why index funds are effective. Being over invested into U.S. equities is a risk if you hold outsized U.S. index funds (esp. if you have a large allocation in the S&P500, you do not own the market portfolio), but there are other risks being invested into foreign equities as a U.S. investor.

Where did I say that? It’s ultimately up to a jury to decide whether lethal force was justified. Obstruct and provoke law enforcement at your own peril.

“Not listening” is really an incredible framing for trying to flee being detained for obstruction, and in the process hitting and nearly running over a federal agent in your SUV.

This will remain the case until we have another transformer-level leap in ML technology. I don’t expect such an advancement to be openly published when it is discovered.

There is a fan project that has revived the servers. It’s hard to get full matches but can still be played at least. It’s called BLRevive, there’s a Discord for it. IMO the only bad thing is it’s based on the post-PS4 update where the devs console-fied the game.

They make great glassware, very scratch and chip resistant.

I.e. disassociating from those people? Isn’t that what homeschooling does inherently? It’s more likely that kids will pick up bad behaviors than they will learn to “deal with” those kinds of people.

I doubt they have offline access to the model, i.e. the prompts are sent to the model provider.

Depends on the distribution of accidents and the distribution of costs. If P(ddos) * Cost(ddos) < P(no ddos) * P(cloudflare outage) * Cost(cloudflare outage) then you would be better off not using Cloudflare.

This is not considering other issues with Cloudflare, like them MITM the entire internet and effectively being an unregulated internet gatekeeper.

You can create macro functions per generic function so something like Vector_New(int)(&v) expands to Vector_New_int(&v). It also looks less foreign (more like templates) than the G macro.