The claim that large language models would replace software engineers this year is aging very badly.

Indeed. If you have a link, I'll bookmark it.

Have "AGI" outperform truck drivers first. They said autonomous trucks would replace all truck drivers by 2018.

Who claimed management consultants are not stochastic parrots?

The software engineers working on the best projects don't follow "scrum" frameworks. Where are the story velocity ticket points in Linux kernel development?

LLM engineers rushing to hard code B is A in their "AI". They call it reinforcement learning with human feedback.

What do you do in your brain when you get something wrong?

Learn the concept better, not an exception to an incorrect rule. It seems like the model architecture just isn't good enough.

Sounds like learning irregular verb forms, or integration rules!

Hardcoding natural language rules doesn't work, we know that from decades of trying. Therefore hardcoding these natural language rules in the model wont work either, it will fail for the same reason, you can't encode enough rules to handle things as well as a human.

Are we arguing ChatGPT is not AGI? I can agree with that.

As for encoding enough rules to handle things as well as a human, what things? Humans don't work with infinite knowledge so there must be a set of rules (however large) that would be "enough".

They don't need to. If you read the thing, you'd realize this was about training not runtime.

The LLM can successfully answer the question at inference.

You need business with them to browse the internet and now to even buy a car. There is a word for this.

I don't see the risk. It seems German companies are making weak offers with "probation periods" even for international hires. One problem might be that such offers are not very attractive.

The probation period is pretty standard. Once you're through it you become much harder to fire, and your job becomes quite secure.

It's pretty bold to give a trial period to an international hire. Employment security is a good thing, and since wages are much lower than in the US, there is not much risk.

Every country does it, except countries with no job security. The notice period of two weeks is more than Americans and Canadians give as a courtesy. What is the alternative, hiring everyone forever?

You don't seem to understand German work contracts well if you compare them to temp contracts.

The probation period also applies to you, the employee, if you choose that a company is not for you. Once that period is over, a three month notice period is quite common, so both you and the employer are locked into a relationship.

The probation period balances out how stable employment is once you both agreed that you're a match.

You don't seem to understand them. German law does not force you to have a three month notice period at the start of your employment, you can negotiate for a shorter period without the at-will part.

You're not wrong. Applied ML articles are not worth reading.

I wouldn’t go this far, applied ML articles are my favorite articles. If you’re in the arena, it’s good to see things that other people have done from a practical perspective so you can ape it in your own work or not give it further consideration.

They don't help you do that, even though poor non-technical managers might think they do. That's why good software projects don't use them. I don't see any story ticket velocity points used in Linux kernel development.

You can't estimate non-trivial software, and if you are doing trivial predictable work, you should be automating it, not endlessly estimating it.

I didn't say I would "deliver" anything. It's done when it's done. You can't predict the outcome of a software project. Many of them fail and micromanagement makes them more likely to fail.

> You can't estimate non-trivial software, and if you are doing trivial predictable work, you should be automating it, not endlessly estimating it.

so if you automate trivial predictable work (for example by making a CRUD app generator), you're moving to non-trivial software territory with costs unknown (as you said, you can't estimate), potentially very high

Or you don't and end up making a non-functional website for the public sector for $100M. Software projects fail often, and not for the lack of trying, there is not yet any evidence that you can micromanage them into succeeding.

Then get rid of the "points" and maybe the "tickets" too.

In my team, points and tickets are useful for knowing we’ve scheduled a solid - not overwhelming or underwhelming - amount of work.

Get rid of the scheduling too if you're doing it too often. Two times a year is good enough.