yorwba

13 ms·

241.
by yorwba 4mo ago
The OSM tag wiki would probably not make for good training data because it only has a single short description for each tag (and even on the French wiki many descriptions seem to be in English?) whereas you want to map multiple descriptions
242.
by yorwba 4mo ago
Students don't like being told they're bad at something. Students don't like being told to do the thing they're bad at. They especially don't like having to constantly repeat this in a loop until they've caught
243.
by yorwba 4mo ago
It's indeed difficult to do good education research, and instructors prefer to just mess around without a control group instead. But the intellectually humble thing to do would be to admit that the experiment can't differentiate b
244.
by yorwba 4mo ago
Worse, because students complained about the difficulty of the AI-graded quizzes, they switch to multiple-choice questions only, which increases engagement, but after analyzing the exam results they determine that multiple-choice question
245.
by yorwba 4mo ago
Those are primarily figureheads with limited power. The EU is not a presidential system. Which is good, because a single person can never well-represent an entire population, directly elected or not. The council is more problematic, since a
246.
by yorwba 4mo ago
Of course no fortification can withstand overwhelming force indefinitely, but el-Fasher held out 1.5 years while completely surrounded, which isn't too shabby. (Here's a map from a year prior: https://upload.wikimedia.o
247.
by yorwba 4mo ago
If you think Chinese companies always act as a bloc, your mental model needs to get about a billion times more detailed. But in this case just a few details may be enough: There are Chinese AI companies that have released LLMs without publi
248.
Thundersnap 0.01: an undo button for everything(apenwarr.ca)
12 points by yorwba 4mo ago|0 comments
249.
by yorwba 4mo ago
Chinese companies giving away expensive models for free is a symptom of the AI bubble, too. It's not a law of nature that they'll always be able to scrounge up the money for yet another training run.
250.
by yorwba 4mo ago
Contrary to what 'b-kf said, you should not prefix your own content with "Show HN" unless it fits the Show HN rules: https://news.ycombinator.com/showhn.html
251.
by yorwba 4mo ago
If you don't care about the notability filter and independent confirmation provided by secondary sources, you don't need Wikipedia. Just be your own primary source publishing whatever you want! You can even make it a wiki and let
252.
by yorwba 4mo ago
LLMs have a very bloated in-memory representation for text, on the order of megabytes of KV cache per byte of text. Meanwhile, for images a lossy representation is considered acceptable and it only takes up maybe a kilobyte of KV cache per
253.
by yorwba 4mo ago
The size is indeed smaller, because text tokens and image tokens are embedded as vectors of the same size, but text tokens typically only cover a few characters, while image tokens typically cover many pixels, so many that you can fit more
254.
by yorwba 4mo ago
They offer a service to let them block sites for you if you want: https://www.ncsc.gov.uk/information/pdns
255.
by yorwba 4mo ago
Most of the fine distinctions are already lost when a document is processed through a pile of linear algebra to turn it into a fixed-size list of floating-point numbers, as you can see from the NDCG@10. Vector search is not a tool for fine
256.
by yorwba 4mo ago
1. Stop worrying about karma. It's not a measure of your worth as a person or something. 2. Be helpful. Answer questions people have. Share useful information that others didn't know to ask for.
257.
by yorwba 4mo ago
Humans may crave newness, but it only needs to be new to them and also still mostly recognizably familiar. So artists who spent too much time steeped in the output of prior artists ended up with Dadaist literature and abstract paintings, wh
258.
by yorwba 4mo ago
By the terms of service: https://www.anthropic.com/supported-countries
259.
by yorwba 4mo ago
You're missing the second-chance pool https://news.ycombinator.com/pool which allows certain posts to reappear as if they were new.
260.
by yorwba 4mo ago
Baidu has an actual AI lab: https://huggingface.co/baidu So does Xiaohongshu: https://huggingface.co/rednote-hilab Pretty much every Chinese internet company seems to have an AI team nowadays, however small
261.
by yorwba 4mo ago
There are AFAIK no plans to completely remove the offline ID everyone is currently required to carry, so I doubt there will be similar rulings for EUDI as long as it remains an optional alternative for people who want to use online services
262.
by yorwba 4mo ago
Renting a venue for an event usually comes with the right to decide who may attend and who may not. So if the embassy indeed rented the park, then as soon as the ambassador decided the journalists weren't welcome, they were no longer a
263.
by yorwba 4mo ago
Meituan published LongCat Flash last year: https://huggingface.co/meituan-longcat/LongCat-Flash-Chat So their track record seems non-scammy so far. Unless you refer to their track record as a food-delivery company and
264.
by yorwba 4mo ago
A per-package charge would have the obvious loophole that you could put many small packages inside a bigger package, pay only once, and then split it up again to send the contents onward to individual recipients. This is already done to sav
265.
by yorwba 4mo ago
Anthropic almost certainly also has optimized software down to the assembly level, considering this take-home interview challenge they published: https://github.com/anthropics/original_performance_takehome/... whi
266.
by yorwba 4mo ago
The primary benefit of releasing weights is the attention it generates. Some people have the hardware to run it, try it out because it's free, tell everyone about it, and then even people who don't have the hardware might get inte
267.
by yorwba 4mo ago
The amount of data Anthropic has claimed was extracted for distillation is tiny in comparison to the entire internet, which is right there for the taking and holds most of the knowledge people expect models to have. Distilling even with sma
268.
by yorwba 4mo ago
That is about investor confidence, not company performance. The companies are for the most part still making boatloads of money, just not as much as investors naively expected to get from taking over a market with 1.4B people. And even if f
269.
by yorwba 4mo ago
What is anti-democratic about trying to assemble a coalition to change lawmakers' minds and get them to vote differently? It is rather core to democracies that decisions aren't set in stone and if you disagree with the majority yo
270.
by yorwba 4mo ago
According to Table 6, it's supposed to be 47.9 to 55.

More ›