Description of the term: https://x.com/karpathy/status/...
About him: https://en.wikipedia.org/wiki/...
I suggest that people go ahead and learn how to use it or be left behind.
I’ve stopped arguing with people about it (mostly haha) for this reason. They’re either going to have to figure it soon enough without our help, or as you say be left behind.
4. Writing the first core version of a service or UI. I’ll typically use close to 100% of those generated lines, and then continue building with LLM assistance where it makes sense. It makes a big difference to development velocity.
5. Finding bugs. If some bug isn’t obvious to me, provide the code to an LLM and describe the problem. Its success rate is high.
6. Working with tech I’m not particularly familiar with (an extension of your #3, i.e. learning)
7. Writing documentation.
8. Reverse engineering existing code, i.e. describe some code to me so I don’t have to dig through it in detail.
9. Writing unit tests.
Her attitude, and loud mouth - not so much.
What do you object to, her support for equality or her opposition to genocide? Or is it that a woman stated an opinion?
LLMs can perform tasks that look like reasoning
So, the functional equivalent of reasoning?
This is why code generating LLMs need to make heavy use of external tools.
Are you saying that ChatGPT, Claude, Deepseek etc. “make heavy use of external tools” to write code? Because they all write pretty good code, up to a certain size of program. Certainly far better than the average human, who can’t code at all; or the average software developer, who isn’t really very good.
I'm not saying I know the answer, but a lot of the content on politically-aligned sites like Parler, Breitbart, Epoch Times, and
Hate speech has been defined in various ways by various groups. Google is pointing me to the UN, which defines it as “any kind of communication in speech, writing or behaviour, that attacks or uses pejorative or discriminatory language with reference to a person or a group on the basis of who they are, in other words, based on their religion, ethnicity, nationality, race, colour, descent, gender or other identity factor.”
There's still a lot of wiggle-room there, and much more grey surrounding what constitutes disinformation. In any case, it sounds like Parler is doubling-down on their mission to enable manipulators and make the radicalization of people as easy as possible.
All the euphemisms in Abode's ad/blog make me nauseous. And for good reason!
Examples: "Return free-loaders to available market" (from TFA), "optimal monetization" (instead of "maximized subscription revenue"), "optimize actions," etc.
This kind of greed-based optimization usually serves short-term goals for a company while sacrificing long-term viability. Instead, maximizing the value provided to customers by providing good content, reliable service, helpful features, and low prices, would help companies like Netflix maintain their strong market position. Customers are increasingly fickle, and the landscape of streaming services is constantly changing. Seeking short-term revenue growth would be a great way for Netflix shareholders to pump & dump their company.
I guess this is the main problem with PoW systems. If mining consumes so much power (and/or dedicated hardware) that very few "average Joes" want to get involved, then the whole point is somewhat defeated. The system is meant to be robust against small numbers of bad actors, but the total number of actors overall is kept small by the value prospect of mining. Thus the move to a proof-of-stake system? Does PoS make small-scale mining profitable enough? I'd expect a reward system to be something like ln(W) or sqrt(W) where W is the transaction-validation work done.
It's the "doubled-down after criticism" part that gets me. Have you ever worked with someone who does that? I have, and each time it was a disaster. It must be some kind of ultra-self-empowerment ideology, pitched as advice for management: Every problem can be boiled down to an issue of perception, and the self-made man owns all of their problems. Thus being wrong is personal failure, a weakness. I don't know, but something like that anyway. Trump is an archetype of this mentality, but of course it's not just him.
As a scientist-turned-engineer, I'm wrong a lot. If I'm not making mistakes. identifying, and then correcting them, then I must not be in a productive mode. Doubling-down on my own mistakes would be self-destructive.
Is this "being wrong is only a problem of perception" BS something taught in ivy league business admin programs, or is it charlatan self-help advice? Where does this come from?
This may be true of a fresh, clean Windows 10 install with almost no other software loaded. But after installing a half-dozen programs or more, and after maybe a week of use, it's back to the usual suckage we all know and expect. So many things scale with installs in Windows that do not in Linux. Registry size, filesystem GB used, desktop icons, and more I'm sure, perhaps update history or app data -- these all seem to load Windows down with O(N) or worse even when the data and apps are not in use. The registry is a database with severely-worsening performance as it fills up. And file indexers and antivirus scans sometimes make a system unusable. There may be something kooky it's doing with interrupts or scheduling, since often any disk activity creates mini lock-ups or extreme unresponsiveness. That makes Windows almost unusable on magnetic hard drives, but much better on SSDs. Still agony, but not completely intolerable once you've given it half an hour or so to quiet down.
It's nice to see here that I'm not the only person who looks at happy Windows users, scratches their heads and wonders "Why? How even?"
This article brings me back to thinking about choosing government representatives by lottery -- sortition.
Sortition probably works best with short terms in office (2-4 years max I'd think), and by removing the popularity contest it should also lessen the effect of power damping out empathy.
My main thesis here is that a random citizen would, on average, be a better representative than our elected reps. I'm thinking more about congress here than president. A president has more need of charisma than average, but on the other hand their role was intended to be that of unification and communication. So they also have great need of humility and empathy. By my standards here, it is simply not possible to elect a good president.
But I won't even pretend to know how to make sortition work in practice. There has to be reasonable compensation for the disruption in one's life. There has to be a short preparatory period. There have to be some loose eligibility requirements. And there has to be a way of declining office. That and fairly strict anti-bribery measures. I don't know what else.
Testing can show the presense of bugs, but not their absence. -- Dijkstra