Posts by adsp42

Arm rolls its own 136-core AGI CPU to chase AI hype train

300W CPU

What's the big deal ?!

My toaster can do 1200W.

Most chatbots will help plan school shootings and other violence, study shows

Loss of internet services

> state progressively taking down 95% of your access to internet services and tech over the next few years, in the name of health and safety.

What "internet" do you have in mind?

The Tim Berners-Lee free internet or the enshitified internet we have today?

I think we sadly have already lost the former.

Your datacenter's power architecture called. It's not happy

Optical interconnects? Re: These power densities ...optical interconnects?

I'm not sure you realise how many GPU boards we are talking about here.

And they all take allot allot allot... of power simply to do the stuff the billion parameters ANN are requested to do. It's not that much about connectivity (between boards, in the rack) but rather the boards themselves.

Software engineer reveals the dirty little secret about AI coding assistants: They don't save much time

Re: circular lists...

Not sure how this is relevant to the article...

In DSP they are a most useful tool. We call them circular buffers.

Google offers 7th-gen Ironwood TPUs for AI, with AI-inspired comparisons

ANN using low-res data formats

> FP8 is an irrelevancy

Possibly for most real world tasks, not for neural networks.

Lots of 4-bit operations too.

AI benchmarks are a bad joke – and LLM makers are the ones laughing

AGI definition

You will have to have another look,

> AGI – vaguely defined by OpenAI as "AI systems that are generally smarter than humans"

That "defined" in the article is a link to a 2023 OpenAI page. They used to think of AGI like that, now they define AGI as

"highly autonomous systems that outperform humans at most economically valuable work"

measuring compute in megaWatts

I beg your pardon. You will find that compute is measured in GigaWatts.

OpenAI non-profit will run for-profit that has yet to make a profit

Re: Silence before the storm?

I must confess I first thought this is a fake, perhaps even generated by an LLM.

I'm more convinced now that it is a genuine academic paper. Still diggestible by general public.

I liked this:

"The problem here isn’t that large language models hallucinate, lie, or misrepresent the world in some way. It’s that they are not designed to represent the world at all; instead, they are designed to convey convincing lines of text. So when they are provided with a database of some sort, they use this, in one way or another, to make their responses more convincing. But they are not in any real way attempting to convey or transmit the information in the database. As Chirag Shah and Emily Bender put it: “Nothing in the design of language models (whose training task is to predict words given context) is actually designed to handle arithmetic, temporal reasoning, etc. To the extent that they sometimes get the right answer to such questions is only because they happened to synthesize relevant strings out of what was in their training data. No reasoning is involved […] Similarly, language models are prone to making stuff up […] because they are not designed to express some underlying set of information in natural language; they are only manipulating the form of language” (Shah & Bender, 2022). These models aren’t designed to transmit information, so we shouldn’t be too surprised when their assertions turn out to be false."

Almost but not quite AGI

"How OpenAI intends to define "most economically valuable work" and how an independent expert panel will verify an AGI declaration is an exercise left to the reader's imagination."

I'm not an accountant, so I'm not sure what roundings are acceptable by trade, I suspect "most economically valuable" must have a finite number of digits after the decimal point.

OpenAI's GPT-5 is here with up to 80% fewer hallucinations

Re: 80% less hallucinations really?

> still get the number of "r"'s in "raspberry" wrong?

Should have gone to Specsavers.

Apple AI boffins puncture AGI hype as reasoning models flail on complex planning

Re: Philosophy

I chat therefore I am

(intelligent) well according to Turing.

Come to think of it, AFI - Artificial Fake Intelligence

Some signs of AI model collapse begin to reveal themselves

Re: What does AI learn?

I keep saying this... The sooner we will start calling it LLM and not AI the better.

Because AI is what we need, but all we get is this LLM cr4p.

Generative AI is not replacing jobs or hurting wages at all, economists claim

Re: Productivity numbers have a problem

Re: Boeing's aircraft

I'd say it's a coincidence.

It's just the rite of passage, every successful capitalist corporation goes. The fat cats are bleeding it to death until the planes fall out of the sky.

Google admits depreciation costs are soaring amid furious bit barn build

Re: There is no business case for AI.

I think many people confuse / equivalate AI with LLM.

There is no business case for LLM. Well, other than a glorified search. Clever interface with internet / computers.

AI will survive another winter, we still have tulips, don't we?

Ireland opens probe into Musk’s X over Grok’s AI data slurp

1 Vote down:

>> If you don't want people or AI to learn from data, don't make it public.

There is a difference between "learn" and make shedloads of money.

2 Vote up:

Indeed "anyone else's AI" is just the same. Only Sam A. didn't dare the Heil salute.

The Reg translates the letter in which Oracle kinda-sorta tells customers it was pwned

Fake news!

Isn't that in line with this new trend in politics to lie people in their face?

Nvidia GPU roadmap confirms it: Moore’s Law is dead and buried

Re: Not Enough Compute Speed?

I think you are missing the point. The Si/Ge/whatever is burning hot and the planet with them.

We are going again the Intel (bust) way of brick_Pentium_with_huge_heat_sink_and_fan_cooler so that we keep up appearances for Moore's law.

Re: Heating the air for telling us bad stories

Don't you worry :⁠-⁠)

When the third (or fourth?) neural networks winter will come

https://en.wikipedia.org/wiki/AI_winter

we'll all be digging more crypto.

Show top LLMs some code and they'll merrily add in the bugs they saw in training

Re: Obvious result, bad methodology

You're right. Just a bit pedantic and missing the point.

They call this "intelligence" when it's just a glorified parrot.

> "Fix the bugs in this code" and it is likely to succeed.

It takes real intelligence to decide that the code needs fixing.

likely to succeed... or to fail. It takes intelligence to decide which one it did, so what's the point?

Re: The clue is in the name

Yeah... I think you've just proven that (most) humans are not very intelligent :⁠-⁠)

> But if we carry on expecting some kind of infallible superintelligent all-knowing artificial being, I think we're going to be disappointed.

... and that God doesn't exist.

Nvidia won the AI training race, but inference is still anyone's game

Re: Need for speed

I'm with you, but let's not forget that internet and social media in particular is full of trivia. And cats. Not that many "solid deep questions" so we are ignored by the profit seeking AgI propagandists.

Very good point Occam's razor, not that easy to use for quality control though.