verytrivial

"drone delivery and humanitarian aid". Right. Thousands of military FPV drone operators beg to differ.

The problem for me is the fusing of the browser with a preferred 'platform' of services. I don't want a partisan browser.

I stopped using Chrome when they started doing the "logged in to Chrome" thing for all Google services. It seemed likely a creepy step in a vaguely defined, unknown direction. The signal seems stronger now.

They're hardware projections into your living space of a massive system run by Amazon. It's the massive system that open-source will have trouble replicating.

I'm going to miss his waffle-free 'Hello' intro. It is/was like stepping on to an airport travelator that's moving 30% faster than you were expecting.

I played this game long enough ago that I needed the LOD adjustment to hit 5fps. Nice write-up!

Aside: I would pay real currency for an official Brøderbund t-shirt. Probably not possible now. I always knew I would enjoy whatever game it was I was starting when I saw that ship logo and the 'funny O'.

I had 2.5 Flash refuse to summarise a URL that had today's date encoded in it because "That web page is from the future so may not exist yet or may be missing" or something like that. Amusing.

2.5 Pro went ahead and summarized it (but completely ignored a # reference so summarised the wrong section of a multi-topic page, but that's a different problem.)

Well, it is perhaps frequently suggested by those Ai firms raising capital that once one of the Ai companies reaches an AGI threshhold ... It as rallying call. "Place your bets, gentlemen!"

I had to take a few runs at that domain name before I could parse it!

Do please take a moment to consider which MPs carry the burden here. It's mainly a single flavour. Mention it on the doorstep next time.

https://votes.parliament.uk/votes/commons/division/1926

I read the opening sentence as "I love DNS erotica" which indicates I've been inside too long and should go for a walk.

That still 8,729 too many sold.

The overlap in concepts required to understand Bloom filters, sets and hash tables is about 95% IMHO. A set is a hash table used for membership tests where you only care about the key, not the value. And a Bloom filter is just a set that exploits the fact that many-to-one hashing 'compresses' the key-space with collisions. It deliberately uses a very collide-y hash function. If a specific key was ever hashed, you WILL get a hit, but there might be other keys that produced the same hash. It's a feature, not a bug.

Thanks. That's a perma-ban for that company from me.

The video for the Dua Lipa track "Blow Your Mind (Mwah)" was shot on the estate for anyone interested in listening to that 2017 banger again.

Sorry to join the pile-on, but can I just ask: In what way does a brain think that an Ai does not? And does the distinction apply from human brains down to fruit flies? Is it a property of embodiment? (I have suspected for years that consciousness isn't just emergent but specifically that it is NOTHING besides that. It's all about scale and large models are just starting climb the ladder. The ladder does not necessarily go up the same way as embodied thought though.)

I thought the comment was about resolving the issue one way or the other without it becoming yet another polarization topic . It probably matters less in either resolution than the cost due schisms and distrust the "debate" causes.

I switched away from Chrome a few years ago when it became slightly evil regarding trying to own login information. I seem to have made the right choice. (I'm not mentioning what I've switched too because it never results in meaningful discussion.)

At least the trains will run on time.

I for one hope this all ends really, really badly for him and his backers.

The only thing that really annoys me about this is that the resources get scraped more than once. Sure, update a delta every month or whatever, but can't the AI bros get the sh*t together and share scrapes? It's embarrassing.

The special sauce is in parsing, tokenizing, enriching etc. There is no value in re-scraping, and massive cost, right?

Just do it.

That Musk is dead.

Are they claiming these robots are also silent? They seem to have "crinkle" sounds handling packaging, which if added in post seems needlessly smoke-and-mirror for what was a very impressive demonstration (of robots impersonating an extreme stoned human.)

Well. https://www.tomshardware.com/tech-industry/artificial-intell...

Our code base still has ghost comments about code being just so because the NeXT compiler won't accept it any other way. No one has the heart to remove them.

I'm going to buck the trend and say this is an excellent idea, but it should absolutely NOT be used for its indended purpose. If OP wants to track his own insulin levels, fine, but to put DIY, uncertified, learn-as-you-go medical monitoring device on someone else, let alone your own son is f*cking bonkers.

Maybe. They're under no obligation to tell the truth about this.

Hypothetical: Take a large short position on NVDA, announce the market that you trained a massive model without using 10s of millions of rare-as-hen's-teeth NVDA-sentsitive resources. Settle postition then quietly settle giant compute bill. Difficult to know either way, but the market seems have taken the team at face value. I guess we'll know if and only if this reduced training cost methodology is replicated.

Is there evidence for the claim that the training was that much cheaper? Or is is 'apparent' by inspection of the verified ingestion architecture?

I can't seem to find any evience beyond the team's statements. What have I missed?

You gain in potential parallelism with FPGA, so with very small "at the edge" models they could speed things up, right? But the models are always going to be large, so memory bandwidth is going to be a bottle neck unless some v fancy FPGA memory "fabric" is possible. Perhaps for extremely low latency classification tasks? I'm having trouble picturing that application though.

The code itself is surprisingly small/tight. I'm been playing with llama.cpp for the last few days. The CPU only archive is like 8Mb on gitlab, and there is no memory allocation during run time. My ancient laptop (as in 2014!) is sweating but producing spookily good output with quantized 7B models.

(I'm mainly commenting to have someone correct me, by the way, since I'm interested in this question too!)

I'm not even complaining anymore. I'm just ceasing to use this Ai garbage and ceasing to deal with people who spruik it. And if you're wondering what "spruik" means, Google will give you an Ai summary which is subtly wrong and contradicted by the first actual search result it list directly below it.