another_twist

I have had to write a Result abstraction multiple times. Honestly I think it should be part of standard java much like Optional.

Whats the problem ? Is throwing directly better here ? Given the annotations I presume this is spring so it will probably be logged anyway. So maybe its unnecessary here.

I mean the "S3" could be replaced with object storage. I guess thats the technical term anyway. Having said that just goes to show how cheap S3 is, if after all of this, the savings are just $500k. Definitely money saved but not a lot.

Use any proper ASR service that supports custom vocabulary ? Transcribe and Deepgram definitely support it and if you want to go fancy Nemo with custom vocabulary.

Are there constraints where you have to use Gemini ?

I guess the opposite, he's probably consolidating power, will likely cash his cheques and fail upwards for putting up a great fight while Meta's engineering culture ends up thoroughly buggered.

This feels overly critical. I think OpenAI definitely shipped quality stuff. It just seems like AI is not the crazy high margin engine everybody thought it would be. Or perhaps that for now, OpenAI is going the Google way of throwing a lot of stuff and checking what sticks. Not a bad way of executing tbh, atleast they're shipping.

These layoffs are in FAIR, not MSL. This is basically poaching researchers over to a different department so they dont have to report to LeCun.

Given that MSL is more product oriented, lets see how it goes.

This is a good step, but it would help to benchmark the project against S3 and its capabilities. Of course it wont be at parity, but it will be useful for people to plan their workloads around this project.

> better for everything else. That means for me: non-coding CLI ops, git ops, writing code with it as a pair programmer, OOD tasks, big new chunks of functionality that are highly conceptual, architectural discussion..

I would argue this is the wrong way of using these tools. Writing out a defined plan in plain english and then have codex / claude write it out is better since that way we understand the intention. You can always have codex come up with an abstract plan first, iterate on it and then implement. Kind of like how we would implement software in real life.

Honestly I think the simplicity of codex to not do anything fancy pants like background coding is what gives it an edge. I am happy to wait for a while and even to repeat context to it (helps me remember stuff anyway) if it types out the right thing.

Regular codex user. Its my typing assistant. Allows me to be the ideas guy when writing software. Codex makes plenty of mistakes when generating large blocks of code but its easier to cleanup and consolidate with a refactoring pass once the typing had been done.

Good programmers worry about the algorithms. Great ones worry about the data structures and the relationships between them. If memory serves, it was Kernighan.

I am so happy something like this is making the front page. Its a great thing.

Its weird that this makes the front page and Metas code world model never did.

So I have read the MIT paper and the methodology as well as the conclusions are just something else.

For example, the number comes from perceived successes and failures and not actual measurements. The customer conclusions are also - it doesnt improve or it doesnt remember. Literally buying into the hype of recursive self improvement and completely oblivious to the fact that API dont control model weights and such cant do much self improvement besides writing more CRUD layers. The other complaints are about integrations which are totally valid. But in industries which still run windows XYZ without any API platforms so thats not going away in those cases.

Point being, if the paper itself is not very good discourse just a well marketed punditry, why should we discuss on the 5% number. It makes no sense.

Right. And why is that ?

This is not a prank right ? Is it really gone ?

I mean honestly my question is why should I bother ?

Java has memory safety and near native performance, the only thing going on for Rust is fearless concurrency which is a USP I think but not enough for me to bother learning its complicated syntax. Had I been unbelievably productive in that language, that would have sealed the deal completely. But alas.

The way I use AI for coding is that I type in the business logic in english and have AI take over the typing. Sometimes I fill in a skeleton and have AI write the boring stuff. But I agree with the author - outsource labor never knowledge. I found AI actually made me a better programmer (judging by feels not metrics) because I know have to explain my thinking which clears up the mud in my head. After which writing the code is leg work better outsourced.

I guess one flaw in the argument about success leading to failure due to model providers eating the product layer, esp for B2B, is it ignores switching costs. B2B integrations such as Glean or Abridge which work with an existing infrastructure setup are hard to throw away and there's little incentive to do so. So in that sense, I dont think AI providers will manage to eat this layer completely without bloating themselves to an unmanageable degree. As an analogy, while Google / Apple control the entire mobile ecosystem, they dont make the most valuable apps. Case in point, gaming apps such as Fortnite who have made billions in microtransactions while running on platforms controlled by other behemoths. They are good investments too.

None of it is information just a sharing of ideas. I don't see a problem here tbf.

I think the author probably solved the blank page problem for me. I dont journal and I have no intention of picking it up. But the trick I picked up here is when faced with a blank page try to treat it as an answer to a question. Solid idea. Thank you author.

Yeah that bit of backstory is something I dont like personally but I get why its written in a motivation then exposition style. People do ask themselves is reading the rest of it worth it.

Quite right. But antlr is better for query parsing. They also have error listeners so error handling can be added.

Why -ng ? I thought it had something to do with angular.

That was because the parent complained about not having good write ups. You can use GPT which has already been trained on publicly available solutions to generate a very good explanation. Like a coaching buddy. Keeping in mind there are paid solutions that charge 15k USD for this type of thing, being able to upskill at just 20bucks a month is an absolute steal.

No its not a measure of cleverness. Its about whether you can break down problems and apply common methods. Thats the entire job. Its a learnable skill and honestly resisting learning because of personal biases is a red flag in my book.

You can make up API calls which you can say you'd implement later. As long as these are not tricky blocks, you'll be fine.

> My biggest problem with leetcode type questions is that you can't ask clarifying questions.

Yeah this one confused me. Not asking clarifying questions is one of the sureshot ways of failing an interview. Kudos if the candidates ask something that the interviewers havent thought of, although its rare as most problems go through a vetting process (along with leak detection).

Hiring people for the test is only valid for online assessment. For an onsite, its very obvious if the candidates have cheated on the OA. I've been on the other side and its transparent.

> It is akin to deciding if someone can write a novel from reading a single sentence.

For most decent companies, the hiring process involves multiple rounds of these challenges along with system designs. So its like judging writing ability by having candidates actually write and come up with sample plots. Not a bad test.