abc-1

That’s mental illness, doesn’t mean being a well paid AI researcher is equivalent to fighting in WW2, sorry.

> Doing AI research at the moment can feel like participating in a war. And from Adolph Hitler to Dutch Schulz. it is widely known that going to war can lead to grave outcomes including psychopathy, divorce and suicide.

Brother… Please… you’re an AI researcher probably making half a million dollars a year to screw around with hyperparams or cuda kernels, shoveling some of the worst python code I’ve ever seen. Get over it.

Great, so there’s no way to “fail” and they can keep asking for money indefinitely. See the problem here? You’d stop shoveling money into research that hadn’t yielded any results either, especially if it was $24 million dollars that could have gone to better avenues.

That was stated as the main goal of the project: improve quality of life in dogs and hope the results transfer to humans. It didn’t achieve that goal so far. I really don’t know what you want me to say.

Nothing useful to improving human quality of life.

Last I heard this had tens of millions of dollars allocated to it and yielded little to no results, unfortunately. Is that not the case?

> OpenAI's engineers: brilliant, A+, A* even.

Then why do they look about average to Anthropic, Mistral, DeepSeek, and many others of that cohort despite having x100 the amount of resources?

Nope nothing wrong about it at all. Some company’s may structure their business in certain ways to take various trade offs, exactly how software devs make certain trade offs. Then of course there’s shady and illegal stuff you’re getting at, but that’s a separate topic.

Fun for a wealthy hobbyist, but if you want to do real work, you’re better off renting from Runpod. Good blog though.

> Our industry is one where actual skills should and do matter [...] Professional rote learning is great for mandarin jobs where you are working within a static prescribed framework (legal, accounting, building codes).

Saying things like this reflects poorly on our community and demonstrates a poor understanding on how much creativity and thought goes into legal and accounting. There's a reason there is a large pay band for lawyers and accountants.

Looks really good, you should be proud.

Haha good one. Do healthcare CEOs next.

Plenty of smart and clever people make mistakes all of the time and you’re only oppressing yourself if you think otherwise. For fear of making mistakes is foolish pride.

Has Uncle Bob even written a single line of code in 20 years? And the stuff he did produce 20 years ago wasn’t particularly impressive.

Well put. There’s a reason turnover is so high at these FAANG companies, despite the perks. Amazon is probably the worst of the bunch.

I thought Netflix engineers were the best and could even do mythical leetcode hards. What happened? Why are they paid half a million dollars a year?

> Due to limited seating in New York and Philadelphia, members were allowed only a single admission ticket, and ladies were explicitly not invited. (Boo.)

We get it, different times had different morals and ethics. But I'm sure the author would have been above all that if they were around in those times.

None of that looks like stuff I can try. I guess the GitHub copilot onboarding was pretty good. Gists are pretty good, but I don’t think he made those originally.

> I have shipped a lot of different projects over the last ~10 years in tech. I often get tapped to lead new ones when it’s important to get it right, because I’m good at it.

Why not list some of them that I can actually try out? How should I give this any credence if I cannot evaluate them?

Anyone who followed Deep Learning in the 2010s would have guessed the same thing. Big boom with vision models by adding a lot of layers and data, but eventually there was diminishing returns there too. It’s unsurprising the same would happen with LLMs. I don’t know why people keep expecting anything other than a sigmoid curve. Perhaps they think it’s like Moore’s law but that’s simply not the case in this field.

But that’s fine, LLMs as-is are amazing without being AGI.

> Here’s a classic example: a data type representing the state of an API request, which can be loading, loaded, or in an error state. If it has loaded, it has the expected data associated with it, and if it is an error, it has some descriptive error. A naïve implementation simply puts those all on a single data structure, with booleans for the states and optional value and error fields:

Nope, not naive at all. Not every single thing needs to be a richly expressed type system. I actually don’t like to see a million different types for some trivial state machine. It is appropriate at times and inappropriate at other times.

Not trying to sound like a grumpy old man- but juniors read things like this and then want to make a bunch of complex types to make “errors unrepresentable” when it’s totally unnecessary and complicating in many cases.

Great blog, but reading it gave me deep secondhand anxiety and even exhaustion. It’s strikingly clear this person is a people pleaser to potentially unhealthy levels, maybe due to low self-esteem or something else, and it could be contributing to the depression. Legitimately, therapy might be the right call here. But on the positive side, they do seem like a good person who wants to build good things. That’s commendable.

Beautiful personal site!

I tried to get it to translate a document and it stopped after a few paragraphs and asked if I wanted it to keep going. This is not appropriate for my use case and it kept doing this even though I explicitly told it not to. The old version did not do this.

Is there a paper on this? Why do no other optimizers give good results? Adam requires insane amounts of memory so alternatives would be welcome.

[flagged]

From everything I’ve read, that seems to be the name of the game on anything beyond senior. Try to gather influence. Slap your name on projects that are going well, stay away from projects that are going bad. Say the right keywords that you’re supposed to say. It all seems so tiresome.

Unless you’re a valuable or high clearance entity, all of this stuff seems like adults having a fun pretend make believe time. Like that neighbor in a nice part of town who owns multiple guns and has a security system set up to protect his maybe… $2k worth of jewelry. And if you say stuff like this, there’s always that one guy who chimes in about that one time when it actually happened for realsies and they were so glad they had their twenty layers of protection and boobytraps set up.

Consider how often SQL, a hash implementation, a data compression algorithm, or a standard library changes. Not often, because they are complete systems. If you don’t like them, you don’t change them- you switch to another system. But they can support an infinite variety of use cases. Hopefully that clears it up for you.

Keep fighting the good fight. It’s more satisfying, even if entropy inevitably wins ;)