This problem only exists because of the marketing move to call anything even slightly ML related "AI".
I see it as a double edged sword. People that want the category of AI to succeed can claim a victory when someone uses it to approximate protien folding and invent new drugs. But that also means the entire field is constantly being dragged down by low quality vibe coded sites, slop videos on social media, whatever horrific thing Grok is doing this week, etc.
>This problem only exists because of the marketing move to call anything even slightly ML related "AI".
We need to remember what "Artificial Intelligence" actually means. It refers to the field of research starting in the 1950's developing algorithms related to combinatorial search, planning, and reasoning. Machine Learning isn't AI in the sci-fi movie sense, but it's among the topics you'll find in a textbook like Russell and Norvig.
A problem like protein folding isn't tangentially related to AI, it's at the heart of the kinds of problems the field has been trying to tackle for decades. Yet when there are legitimate breakthroughs, people deride it as "not real AI."
That isn't how we break up the field of AI. ML are the algorithms that are based on statistics and other numeric methods. The other algorithms are based on logic and perhaps some philosophical methodologies. We don't really have a name for that second group, its just part of AI. Then there is Reinforcement Learning which is a sub-field of ML and incorporates a Pavlovian methodology.
AI is the sum of all of these groups. Also, the "not real AI" thing is more about not real AGI. That's a very different target.
I'm not sure if your comment disagrees with mine. I think we agree that machine learning is a sub-field of artificial intelligence. Evidenced by the fact that Russell and Norvig, the most authoritative textbook on the subject, includes multiple chapters on ML.
AGI isn't a well-defined concept. So when people say something isn't "real AI" because it's not AGI, I can't take them seriously because they're implying that everything the field has worked on for the past 70 years isn't real AI.
There's no branch in that code either way. The comparison operator outputs a value (which is arithmetic, not a branch), and that value is added unconditionally.
The check isn't important; what's important is being predictable so the CPU can guess which way the check will go. I don't know exactly how it works, but after the first couple of loops, the predictor will assume it's always going to end up in the loop and make that the fast path. It may guess wrong the first couple of loops, and the last check wrong, but the other 997 will be correct.
There is a static branch predictor that is used if there is no statistic on a branching instruction yet, and it's really simple: Jumps backward are assumed to be taken (they usually are from a loop), jumps forward are assumed to be not taken.
So the jump that forms the loop will be predicted correctly for all executions but the very last (when the loop ends).
Unless the loop is unrolled, yes, there is a branch to exit the loop. But then that doesn’t matter because the whole goal at the beginning was to avoid branch misprediction (which is not the same thing as avoiding branches entirely).
same author who's idea of constraint decoding for structured outputs was to run an schema-begging-API call in a loop 10 times & then throw an exception on failure.
OpenAI per employee valuation is $150M+ (almost 100x of per employee valuation of our company). I think it may make sense to ponder a bit why a $150M engineer would have such an idea. May it be it is the preferred way of doing things in the new AI world, a paradigm shift.
We all know the most highly valued people in the world are the smartest. In fact, the more money you're being paid, the more value you add, the more your opinion is worth, and the more skill you have. The amount of money you're worth is actually directly proportional to your worth as a human being
HTML made by Claude will, by default, be "sleek, modern", with colorful tables, cards, maybe Tailwind for styling. And, of course it will, if you wanted a barebones HTML, you would just have asked for markdown!
So the LLM decided to present some content using 4 cards, and you now want to add new itens. You can't just add new lines of text: you need to copy the whole HTML of the cards. But the LLM used different colors for each card, so now you have the first cards with varying colors and the new cards all the same color as the last card. Now you have to think about colors... etc etc
It depends what we mean I guess, isn’t Markdown supposed to allow [hx]ml tags anyway if user need them? Then it’s more about asking the LLM to generate Markdown with this in consideration, and privilege rendering the output of reports in the preferred browser after relevant rendering.
1. I believe many applications that use markdown allow html. Others don't due to security/rendering issues.
2. One of the limiting factors of LLM is context. An html table takes up way more tokens than a markdown table. Especially if it's a WYSIWYG editor that has all kinds of css and <span> tags just for fun.
> An html table takes up way more tokens than a markdown table
That might be the case today but there’s no reason for it to always be true. They are different representations of the same thing, an LLM could (arguably should!) store an internal representation that uses fewer tokens.
Elsevier is shitty to people doing stuff that (imo) should be allowed. Meta is making money doing the same thing and not getting the same shittiness from Elsevier.
Elsevier at least works within the (admittedly broken) system, Meta does not.
It was better because the Civil Aeronautics Board did not allow price competition, so the cost of economy tickets was much too high, on some routes reaching the inflation-adjusted level of business class tickets today.
As I understand it, everything about the industry was better back then too.
Case in point: Old Perry Mason shows where characters regularly drive to the airport, pay for a ticket and get on a plane. Flying was actually faster than driving back then, even when measured by time between deciding to leave and arriving at destination!
(Yes, tickets used to cost a bit more. Whatever. Figure in the price for camping in the airport for 4-5 hours, and then tell me the current system is cheaper!)
Tickets used to cost 4-8x what they cost now, depending on route. It wasn't a couple percent extra. A lot of what made flying seem like such a glamorous activity was that everyone but the upper classes was excluded.
An economy class round trip from the US to Japan in the 1970s with Pan-Am was $8,900 in 2026 dollars. About $15,000 if you flew first class.
And for comparison, today you can do an economy round-trip flight with Delta Air Lines for roughly $1.6k (SEA-HND). A Delta One flight is roughly $8.5k. That's the apples-to-apples comparison.
Deregulation also allowed international carriers to sell to us too. An ANA round-trip on economy class is a couple hundred dollars cheaper. Their business class is similarly cheaper than Delta One.
Air travel is so much cheaper than it was back then that it is affordable for most people to take one international trip a year if they really want to. Even to exotic places in Asia or Southern Europe.
It would be prohibitively expensive for poor people to fly. I understand why you wouldn’t care about that, but some people are poor and still need to fly if you can believe it.
At least on x86, multiple additions and multiplications can be done with a single `lea` instruction so it's preferable to XOR. Though I have no idea about other architectures, compiler implementations, any interpreters...
That only helps with multiplications by statically known word sizes (4x, 8x, etc.) and not arbitrary x·y. It can help with many smaller constant multipliers if the complete is clever, but it has to be known at compile time.
I see it as a double edged sword. People that want the category of AI to succeed can claim a victory when someone uses it to approximate protien folding and invent new drugs. But that also means the entire field is constantly being dragged down by low quality vibe coded sites, slop videos on social media, whatever horrific thing Grok is doing this week, etc.
reply