blharr

It briefly touches on training, but uses a seemingly misleading statistic that comes from (in reference to GPT-4) extremely smaller models.

This article [1] says that 300 [round-trip] flights are similar to training one AI model. Its reference of an AI model is a study done on 5-year-old models like BERT (110M parameters), Transformer (213M parameters), and GPT-2. Considering that models today may be more than a thousand times larger, this is an incredulous comparison.

Similar to the logic of "1 mile versus 60 miles in a massive cruise ship"... the article seems to be ironically making a very similar mistake.

[1] https://icecat.com/blog/is-ai-truly-a-sustainable-choice/#:~....

Where?

One of the most crucial points "Training an AI model emits as much as 200 plane flights from New York to San Francisco"

This seems to come from this blog https://icecat.com/blog/is-ai-truly-a-sustainable-choice/#:~....

which refers to this article https://www.technologyreview.com/2019/06/06/239031/training-...

which is talking about models like *GPT-2, BERT, and ELMo* -- _5+ year old models_ at this point.

The keystone statement is incredibly vague, and likely misleading. What is "an AI model"? From what I found, this is referring to GPT-2,

It used to save my ass a lot. It still does, but it used to, too.

Such a simple operation, but it'd probably fix 95% of problems I've had with git.

"Uh oh, I tried to checkout a branch from remote by doing `git checkout origin/some-branch` instead of `git checkout some-branch` and made a couple source changes. Now I'm in detached head state. What is detached head ? I have to stash my revisions? Can I make a new branch with the same name or do I need to delete the origin/some-branch that I'm on?"

When you could be able to just revert the "operation" and check out the correct branch, that's amazing.

The biggest thing is to not focus too much on openings. Past 4 or so moves everything can be mixed up and tactics are more important.

You're at the level where you're probably not hanging all your pieces or hanging mate-in-one too commonly. So learn endgames. You can win/draw like 75% of the time from even being down a piece or two at your level if you can play the endgame quickly and accurately.

Then, refine the middlegame, learning how to get the advantage. Doing puzzles for dozens of hours will teach you how to recognize a lot of basic tactics that can win early.

But only once you've scraped the barrel with the middlegame would I start focusing strongly on openings.

If you memorize the documentation I suppose there's no problem with it, but even then there's confusing things from the beginning design approach.

Like, reset vs revert vs restore, using three similar starting synonyms for different operations.

Reset is particularly confusing because it sounds incredibly destructive, but if you do a soft 'git reset', it just moves the "changes to be committed" to be not staged for the commit.

Then, if you change a single flag (--hard) it is destructive, instead erasing all those changes from your disk.

And there's a reason nobody reads the docs, because they're laid out confusingly at first glance to an unfamiliar person.

If I go to git-scm.com, click docs, the first suggestion is the reference manual. The first guide is on "gitattributes" which gives no philosophy or context as to what it is even used for. Ok, maybe, I'll just check the complete list of git commands... and it starts talking about porcelain and plumbing. Nope, thats not what I need. Maybe try the link "git" under Setup and Config. There the description finally links to a useful "gittutorial" page.

The gittutorial even has some confusing or specific terminology that makes git seem hard. From the beginning of importing a new project "Assume you have a tarball project.tar.gz with your initial work"... why do we need to refer to tar at all, and it doesn't even do the credit of explaining the tar xzf command or what that tarball is. It could easily say 'assume you have a directory', which is the more common case and be 200% more simple.

Tornados might be more intense but only for a short period of time and in a small area. I don't see any of those where the tornado is lasting days, causing sustained damage. There are some where there are multiple tornadoes in a span, but each individual tornado is itself quick and violent but localized within a mile or so at most.

Compare some incidents with, Hurricane Sandy, for example, where it traveled across the span of a thousand miles and lasted a week of damages.

I mean, it's a good idea, I just wouldn't buy the one in Korea to get shipped over here. I don't get the cynicism, someone in Korea had this idea and made the product, probably intended for other people in Korea where the shipping isn't an issue?

As a bit minor criticism "Tasks Completed (Time)" is hard to evaluate without the time intervals or units. I'm not sure if it should just be "Time"?

m ∈ ℤ is the threshold for determining high or low novelty

search(T,θ,m) retrieves the first m historical tasks that are semantically similar above the θ threshold

Are both m's here the same or different numbers? I found this a bit confusing

You can find the phi-4 technical report [here](https://www.microsoft.com/en-us/research/uploads/prod/2024/1...)

The brief of it is by curating a smaller synthetic dataset of high quality from textbooks, problem sets, etc. instead of dumping a massive dataset with tons of information.

I mean they have name recognition and a userbase, but they're hardly the best at doing any of those features.

Default destination for many is still just Google, and they've added AI to their searches. AI chat boxes are shoehorned into a ton of applications and at the end of the day it'll go to the most accessible one for people. This is why AI in Windows or in your Web Browser or on your phone is a huge goal.

As far as extra features, chat GPT is a good default, but they're severely lacking compared to most other solutions out there.

Even in America I've seen similar opportunities. Engineering PhDs at state schools have offers where you get paid a small but livable amount and get your tuition paid for. I think the big difference is that in industry you can just get paid way more in America compared to engineering elsewhere.

import os; while True: eval(input())

"They'll just get better" is a huge non sequiter.

Adding autoplay videos that are tangentially related to the wiki page.

Said videos take about 10 seconds to show up and will cause the screen to scroll to a different point from where you are currently reading

Their privacy policy is exhaustive, and you opt in by default to sharing your data with Google and 7 other analytics providers

UI noise like "share to Instagram/YouTube/LinkedIn" that probably 0.1% of people even consider using

When you're on a specific game wiki, you care only about that wiki content, but fandom places a Fandom navigation bar and side menu

Their default email strategy sent almost daily updates for random posts on (unrelated) FANDOM community when I made a couple comments on a specific wiki a decade ago.

Only 40% are obese (US) so it affects a minority of the population, therefore it should be disregarded?

Someone else in this thread said,

> With a free account the model it claims to be using is "4o auto", which is not a model but apparently an attempt to automatically decide models for you to be more cost effective.

It got it right, but an interesting result that it rambled on about monetary value for... no reason.

> While the lint bag is heavier in terms of weight, it's worth mentioning that gold is significantly more valuable per pound compared to lint. This means that even though the lint bag weighs more, the gold bag holds much greater monetary value.

LLMs are probably the worst at programming in Lisp in my experience though

I mean a 30 year old solar panel isn't going to be useful to be fair

The header "Matt LeBlanc" and the photo is the actor who played Joey. They asked the name for a baby kangaroo and got [some] information related to the name Joey

Right. And so you'll end up punishing some sites that do it perfectly, while other sites that game the algorithm game first priority.

I've had the exact same experience, but opposite. Tons of things that are trivial in C# take a ton of code in C++ to me. Maybe it's just going from being an expert in one language to a newer language?

Isn't instant communication impossible, even with entanglement?

I mean, or it heavily restricts the pool of founders to mostly those who already have enough wealth such that they won't be homeless.

I don't see why a safety net wouldn't mitigate this risk in the first place, which would then allow for more people to take these risks and come up with great ideas as well

Couldn't the LLM though just randomly replace/reword things to cover up its frequency in "post"?

I mean, we get it. It's a UX problem. But the thing is you have to tell it exactly what to do every time. Very often, it'll do what you said but not what you meant, and you have to wrestle with it.

You'd have to come up with a pretty exhaustive list of tells. Even sentence structure and mood is sometimes enough, not just the obvious words.

Glad I'm not the only one. I see 4o as a lot more of a sidegrade. At this point I mix them up and I legitimately can't tell, sometimes I get bad responses from 4, sometimes 4o.

Responses from gpt-4 sound more like AI, but I haven't had seemingly as many issues as with 4o.

Also the feature of 4o where it just spits out a ton of information, or rewrites the entire code is frustrating

This discussion includes o1. It's marginal improvement compared to, say, GPT-3 to GPT-4.