You’re absolutely right!
Sorry, couldn’t help myself. Good metrics are hard, but token leaderboards are obviously terrible metrics that will lead to bad business outcomes unless propped up by investor hype. Of course investor hype is a real factor, and it’s often driven by guesses about what other investors will do rather than real business outcomes.
It seems obvious to me that there’s a huge gap between a search result, which is a link and an excerpt where the operator of the linked site is clearly responsible for the content, and an AI overview, which mixes information from multiple sources with generated nonsense.
Google should absolutely be liable if it generates overviews containing defamatory lies.
They weren’t going to let people have it that cheap for very long. The plan was to offer it on subscription plans for a couple weeks, then move to usage-based billing, which is much more expensive for a usage pattern that comes anywhere near the subscription limits.
Keeping a single instance of Fable busy for a full day would probably cost a thousand dollars at standard API rates, and some agentic coding workflows run many agents in parallel. Companies have just recently started to figure out that rewarding employees for how many tokens they use may be a waste of money, but Anthropic is hoping to cash in before they all do.
If it’s true that young software developers are mostly not productive, then employing them hasn’t made sense for decades because staying at the same company for most of one’s career stopped being the default some time around 1980.
The battery law is “so broad and so restrictive that it prevents the sale of this wonderful, jointly developed, U.S.-European product from being sold in the European Union…”
No. Meta chose to design e-waste.
I’m curious as to how people were actually using Google Glass in 2013. The only place I saw one then was at a programming meetup. I’m quite sure people weren’t regularly running the camera continuously because that drains the battery in a hurry and overheats it even faster.
Source: I bought a new old stock Glass XE-C this year. They’re $50 on Ebay. I’m working on a navigation app for it because I want heads-up cycling directions.
How to motivate right to repair advocacy in other regions in one easy step…
Sure, but many languages do that,
I wrote several paragraphs and talked about three languages, so I’m going to have to guess about what “that” refers to. I’m guessing it’s Lisp macros. Your other comment offers template metaprogramming in C++ as an alternative.
Template metaprogramming Gets maybe a third of the way to what Lisp macros offer. It can do compile-time syntax transformations, but it doesn’t provide the full C++ language with which to do so, doesn’t operate on the actual parse tree, and isn’t Turing-complete in practice because of fixed limits on recursion depth in real compilers. Rust macros get much closer, providing the full power of Rust and the option to get at a real AST by parsing the token stream they operate on.
If you mean something else, please elaborate. It’s an interesting topic.
There are more modern and better ways, IMO.
I’m not sure what “more modern” means in this context. If it just means young, I can probably find a Lisp family language with its first release this year, though that wouldn’t be the one I would recommend to a beginner. If it means recently-updated, Racket, the Lisp I recommended learning had its latest stable release nine days ago. If it means something else, please say so.
“Better” probably can’t be measured objectively, but by all means, make the case for something else.
Some people want to learn programming to get a job, though perhaps not as many in 2026. Some people want to do a project that happens to requiring programming. Some actually want to understand programming and get good at it. The last group will benefit from learning Lisp and Haskell even if they don’t end up using those languages much. I thought my first comment explained why and I think Corngood elaborated on it, but I’ll add more.
The reason to use programming languages instead of machine/assembly languages is that they add abstractions, and allow the programmer to add more abstractions. An abstraction is a name and implementation for a repeated pattern in code, which documents the programmer’s intent when it is used, allows all invocations to be modified in one place, and substantially shortens programs. In most languages, there’s a distinction between abstractions the language designer can add and those the programmer can; in Lisp, there is not.
If most languages didn’t have if or class, you couldn’t add them in a library; you’d have to modify the interpreter or compiler. Here’s if defined in a Lisp-like language I’m working on:
(defmacro if (test then else)
`(cond ~test ~then true ~else))
This is possible because Lisp code is made of Lisp data structures which it can easily manipulate, and because it has the ability to control when evaluation occurs. Here, we need to splice three blocks of code into a cond expression, which is a more generalized form of conditional evaluation that takes an unlimited number of test/then pairs. We must also prevent the premature evaluation of the branch not chosen, which is why if and cond can’t be regular functions. In Common Lisp, the entire object system can be implemented as a library.
Haskell and similar languages also offer significant power for abstraction with its sophisticated type system and lazy evaluation, but the more important lesson they can teach is the gurantees they can make at compile time. Once a Haskell program compiles, it has a much greater chance to work as expected than any other language I’ve used.
SQL teaches thinking in data. Most programs exist to store and manipulate data, so that’s pretty relevant.
I question the suggestion that Zig and Go are not “serious” programming languages. They certainly weren’t designed to be “easy” beginner languages.
I don’t think it matters a whole lot which language you start with. Learning to program is largely separate from learning a particular language, and if you do programming for a while, you’ll probably learn several. I do think someone who wants to understand programming deeply should learn each of:
The policy is mandated for “certified” Android devices. Those have Google apps preloaded and pass Google’s remote attestation.
GrapheneOS is not certified and has no plans to be.
If someone was going to reward me for burning a lot of tokens, I’d feed LLM output into the LLM input until they ran out of rewards.
I am surprised. I’ve run both KDE and Gnome on a Surface Go 2 (8100Y), which is either slower or barely faster depending on which CPU you have, and I’ve had no UI lag.
My solution in the same situation is to use Gnome. I strongly prefer KDE on a desktop/laptop, but Gnome is an outstanding tablet UI and KDE isn’t… except that Gnome’s onscreen keyboard is crap.
That means if Google’s verification system gets widely adopted, browsing the web could become a headache.
Using a phone to scan a QR code in order to access a website on my desktop is a headache even if it has no dependencies in particular.
It seems very unlikely to me that the model itself has a list of banned words, and much more likely that a purported list is hallucinated.
If they did want to have a simple list like that, it would probably go in the harness rather than the model, and the model wouldn’t have been trained on it, nor would a reasonably designed harness provide it to the model. Legitimate use cases, such as asking the model for a list of abusive words for use as a first pass in a filtering system could get tripped up.
As a test, I asked Perplexity to generate such a list. It did a bad job, including such words as abuse, hate, and threat which are far more likely to be innocuous than abusive. It did also include some highly offensive slurs that one would expect on any banned words list.
The intention is to mock everything in tech. He mostly mocks big tech, but nothing is immune.
The channel description says
SAMTIME is probably the only channel on the internet not smelling the farts of the big tech companies! From Apple to Samsung, Huawei to OnePlus, we make fun of everyone and tell it like it is (aaand maybe exaggerate just a little if it’s funny).
and that seems about right from the handful of videos I’ve watched. He spends a lot more time mocking big tech firms than poking fun at Linux.
Android was out for at least five years before Safetynet was a thing. I’m surprised people weren’t louder in their objections to that then.
Improved reliability is, ironically a major cause of this problem.
Buyers of brand new cars in the USA keep them for fewer than ten years on average. If most cars can go ten years without needing major repairs, the car manufacturer’s actual customer is not strongly motivated to consider repairability in buying decisions. The second or third owner likely cares a lot, but their preferences don’t matter very much to manufacturers.