I think people who associate themselves with more or less any organization who profits by producing forms of entertainment specifically taking advantage of people's bloodlust are "laying with dogs".

This includes but isn't limited to violent video games and sports.

Though, I see what you did with racism there. Great plug. Poorly placed as the post you responded to didn't appear to make any suggestions of that. Maybe you're a little bit racist by suggesting that people making analogies to dogs must be thinking of a particular skin color. Is this what you do?

OpenAI has burned brilliantly. They put on quite the show. But in the end, they really had no long term product.

OpenAI's had a great head start. And with piles of money, they could burn the GPU hours to stay ahead.

The gap is closing. The market is flooded with unused GPUs. And OpenAI's worst enemy, real programmers.

There is nothing happening in AI that dictates that you need more bandwidth or more cores. The thing is, most model training is run using extremely inefficient code. Memory access is so unoptimized that there are at least several orders of magnitude if CPU cycles wasted on inefficient memory access. We attempt to solve this with faster and lower latency memory. Of course, optimizing the code will have much greater impact and eventually move training to laptops.

And most importantly, Apple, Microsoft and Google will eventually be able to deliver full AI on their own. When this happens, since the platform is the product and AI is just a feature, no one will bother with 3rd party AI.

Feed a more or less blank model with millions or billions of math problems. Then check the answers and keep feeding the problems back to the model... But don't just keep trying the ones the get wrong. Then make an AI generate harder and harder problems and use tools like Wolfram to test the validity of the questions.

The more you feed the model, and refeed in randomized order the more accurate they become.

Wonderfully stated.

I work at the university. My daughter also attends. I pay taxes, taxes pay for school.

It is the government's responsibility to educate the people who will sustain and support it. This is common sense. No government should ever let a child start their life in crushing debt. And any government worth its salt would want their people to be an asset to the country. This means investing in their children.

When I ask Chinese people about the US, they give me responses from sad signs to regrets that they may never see America.

I have never seen met a Chinese person who appeared to have the slightest malicious thoughts towards the US.

When I've asked them about Trump or Biden, some show great dislike towards them. As though they are evil people trying to cause problems.

Food for thought... These people are generally true capitalists. The of course believe in socialism because only fools don't. But they are believers that if you make something cool and then they take the time to copy it and hopefully improve on it, if they advanced and you didn't, that's your fault and if you can't compete, you're doing it wrong and its better for the customer that they take over.

Taiwan is funny. I've asked many Chinese people about it. Inside China, at Huawei headquarters, on planes...
  - Only one reacted strongly... "It's ours".. All he snapped and the conversation ended
  - Older people often say "they'll come home when they're ready"
  - Almost people said "I would never want to fight for it." They would be fighting themselves.
  - Quite a few said "Take Taiwan? Take it where? Isn't just where it's always been?"
  - Some said "I'm from Taiwan. I just moved here for opportunity. "

With regards to governments, both countries are involved and they're fighting an unwinnable trade war.

With regards to the people, it's a very one-sided conflict. America is a bully causing trouble while China waits wondering why their friend is being mean to them.

There is a magical tool called pathfinding.

Cut a deal with google, use Google maps and either brute force shape fitting and pathfinding to route orders to the optimal store or use ML to estimate the solution and avoid travelling salesmen. Then you can even consider traffic conditions as well.

This is not a difficult problem to solve.

I used to hire juniors, train them, build a project and spin them off.

Now, I just use AI and do it myself.

The ship has already sailed.

Intel is impossible to do business with. My department has a budget that if Intel would pursue us would actually make a difference on their quarterly targets. If they actually listened to us, it would buy them at least half a fab. But, even though we're one of the richest and most liked countries (outside of Texas, we're socialists, they don't like us) in the world Intel won't even send one salesperson to us. Every other vendor does and every other vendor gets some of our money.

If they spent a hundred thousand a year to plant a sales rep in our neighborhood of a national capitol, they would have access to $billions of IT budgets. My project which I work on has a $5 billion a year budget. Intel doesn't want any of it.

We have a two hundred million dollar budget for AI accelerators. AMD, NVidia and GraphCore are practically sleeping in tents at our door. We can't even get Intel to pick up the phone.

I offered to evaluate Intel network adapters for a 5000 node supercomputer. They said "we'll call you"... They didn't.

They apparently prefer to spend their time and money complaining that no one wants to buy their stuff.

Every Enterprise HDD contains with a big chunk of flash and often a big chunk of RAM with a "super capacitor" to flush RAM to Flash on power loss. SSD controller chips are also much simpler than HDD.

So since an HDD is basically an SSD plus a whole lot more. I can't see how this could be so.

I listen to a lot of audiobooks which will never become films or if they do will be a disaster on a "what Disney did to Artemis Fowl" scale.

I would love to say "<voice assistant name here> make me a 20 episode, 45 minute per episode series reminiscent of <insert book name here>. Use idealized actors. Limit gratuitous sex and violence. Make the actors hot without sexualizing them. Prefer European fashion over American. Use clear civilized accents and dialects. No nose talkers. Avoid Southern US twangs and drawls. Make the music epic with a hint of metal", and then start watching.

We are rapidly heading there.

I refuse to believe they are actually puzzled by this.

Is it more likely the article author contacted a PR spin doctor who supplied an answer to a question they hadn't been prepared to answer and dropped that answer as a placeholder?

This was an amazing article until some idiot made the moronic remark about AI.

We need 80-128GB of high speed, probably volatile storage to run the biggest LLMs right now. Let's assume eventually a terabyte will be the viable economic sweet spot for functionality vs the buck. The write time has nearly zero effect on "real time". It's all about the read.

So this is fast memory. But once you
1) multiplex it
2) add error checking and correction
3) connect a system bus (assume 20cm)
4) Serialize and deserialize data

How fast is it then?

This is a massive achievement and proves China is heading into the lead as an innovator.

Why ruin the article with nonsense about AI?

It goes up and down.

Can Bezos make it go left and right, maybe forward and back? Possibly shake things up and try to yaw, pitch or roll?

Clearly, this must be within the realm of possibility for The Bezos Eye (like the London Eye but you have to be a great publicity stunt to ride it).

Maybe for this flight, he can send all Africans or maybe for real fun a bunch of flat earthers. Hasn't it been a while since an all dog or primate crew has gone to space?

P.S. My jabbing isn't because I dislike BO. It's just that I'm turning 50 in a few weeks and I worry I won't live long enough to see BO actually advance past anything other than joy rides. It's time for Bezos to buy his own president or pay Trump for a few months of playing overload to fire anyone blocking him from progressing faster.

Is 1.5 million a lot? On a cosmic scale and with enough data, 1.5 million seems pretty insignificant.

Given the little I know of the data set, was this an efficient solution? Did we need ML or could video compression style motion search yielded similar results?

$250,000 sounds like a lot. This really didn't seem all that impressive. Out of the top 300 kids announced in January, was this really among the best? Step 1) take data set and known positives from earlier searches. 2) train model 3) process 4) write presentation. Using modern tooling, this is maybe a day of watching YouTube videos, a day of work, a few days of writing.

I feel there must be a lot missing here

My employer files my earning with the government.
My pensions file with the government.
My banks file with the government
My side hustle soul proprietorship files with the government
My side hustle accounting software detects when I spend money and nags me for the receipt and automatically files it properly
I use 10-20 minutes on taxes a year through a government portal using the standard national multi factor authentication.

I get to spend the rest of my time laughing at Americans who can't even trust their own government.