EternalFury

Everyone is starting to express themselves as LLMs. I fell prey to having LLMs “copyedit” some of my words when I wanted to make sure I got my point across, but I am trying to stop that. Online in particular, I see a style which I call “probabilistic English”, infused with odd hip expressions.

I thought it was only a problem for English communication, for which we have little care, but the same applies to code now. More and more code appears to be “probabilistic code”, homogenized to what is most common in the data LLMs were trained on.

Public education is important. Without it, it’s harder to stay above average. But there are those who say “it is not my duty to fund the education of anyone else.” Hence, here we are.

Yes, it’s contradictory to some degree. On purpose. That’s part of living it.

My “advice” is also US-centric, where I have observed time and again what truly leads to one getting their fair share.

Freelancing is a way to achieve this.

My advice is: Don’t let someone else control your life. Switch occupation every 2 years, even if you find a great environment. Don’t chase, for what you need must come to you. Understand no one can give you the answers you seek; you have to live it.

If somehow recovering the capex expenditure is not counted, if somehow the cost of developing future models is not counted, then yes, inference costs of current leading models allow a profit.

But those things are tied together.

Even xAI, that now has a reasonably competitive model, is struggling to achieve PMF. Meta is in shambles because their models have underperformed for years now.

There are people who think knowledge discovery is just a matter of parroting past behavior and trying things at random until something sticks. I don’t.

Force is supreme until you use it, then everyone knows it has limits.

So…calculators are intelligent? How about accountants that failed arithmetic 101 in high-school, are they intelligent? Generally intelligent?

We are jagged, but we can smooth that jaggedness if we choose to do so. LLMs stay jagged.

The real question is: Can it be generated using programs? If it can be, then LLMs will eventually monkey type these programs.

Yes. I doubt it can do that.

I am thinking there’s a large category of problems that can be solved by resampling existing proofs. It’s the kind of brute force expedition machine can attempt relentlessly where humans would go mad trying. It probably doesn’t really advance the field, but it can turn conjectures into theorems.

It does seem good, but it’s slow.

I was going to say something, then I realized my cynicism is already at maximum.

Is it just me or does it feel like billionaires of that ilk can never go broke no matter how bad their decisions are? The complete shift to the metaverse, the complete shift to LLMs and fat AI glasses, the bullheaded “let’s suck all talents out of the atmosphere” phase and now let’s freeze all hiring. In a handful of years.

And yet, billionaires will remain billionaires. As if there are no consequences for these guys.

Meanwhile I feel another bubble burst coming that will hang everyone else high and dry.

They learn the value of specific actions in specific contexts based on the rewards they received during their play time. Specific actions and specific contexts are not transferable for various reasons. John quoted that varying frame rates and variable latency between action and effect really confuse the models.

100%

What John Carmack is exploring is pretty revealing. Train models to play 2D video games to a superhuman level, then ask them to play a level they have not seen before or another 2D video game they have not seen before. The transfer function is negative. So, in my definition, no intelligence has been developed, only expertise in a narrow set of tasks.

It’s apparently much easier to scare the masses with visions of ASI, than to build a general intelligence that can pick up a new 2D video game faster than a human being.

Such a weird move. They must have too much cash because they are not buying GPUs from Jensen.

It costs someone something, but no one their freedom. Mass ignorance is the opposite.

As for degrees with no use, pretty sure these are the byproducts of education for profit, with heavy marketing passing as administrative expense.

Maybe you could divide the system in two halves: 1) Of national interest, 2) Discretionary.

As for earning potential, it has nothing to do with free education, as so many high-earners in the US were educated by such systems.

If anything in any country should be free, it should be education. And, obviously, the administration of education should never be a for-profit venture.

Valuing democracy and being able to select sensible leaders depends on it.

Yes, you must be able to recognize the right answers and separate them from the hallucinations and crappy engineering. Maybe it’ll get to a point where even people who don’t have a clue can click Apply and never have to worry, but it’s not quite there yet. If it does get there, I’ll find another occupation; it’s not like my job ever defined me as a person.

I have been doing this for 30 years now. The software industry is all about selling variations of the same stuff over and over and over. But in the end, the more software there is out there, the more software is needed. AI might take it over and handle it all, but at some point, it would be cruel to make humans do it.

Isn’t this equivalent to maximizing latent space activation without corrective user input? How does it implement self correction or backtracking?

What’s the time and space complexity of the new approach?

Business as usual. While electricity is remarkable, no one gets extremely rich selling it. End-user value is the only value that can be sold at a profit.

I know. The question is: How much of the Internet trove, including the smart bits, but also the tremendous amount of inane content, is actually useful to building the foundation that allows 1,000 problems to have such an effect?

Just imagine a textbook that gives you the understanding you need to score high in math competitions…and it describes less than 1,000 problems. This in itself is a major discovery in metacognition.

o1-preview, o1, o1-mini, o3-mini, o3-mini (low), o3-mini (medium), o3-mini (high)...

What's next?

o4-mini (wet socks), o5-Eeny-meeny-miny-moe?

I thought they had a product manager over there.

They only need 2 names, right? ChatGPT and o.

ChatGPT-5 and o4 would be next.

This multiplication of the LLM loaves and fishes is kind of silly.

That’s such a show of confidence! It’s an entirely different driving world over there. Particularly in terms of what is expected outside of what is mandated. Looking forward to trying it.