Yeah I find that when people say they are strength training and find it boring, it's usually because they aren't challenging themselves with heavy weight.
I agree. Loading up a bar you're not sure you can lift... staring at it, prepping form, breathing deep, planning your move, then pushing as hard as you - safely - can... it's very engaging.
My problem is with the rest period, not the lifting part. I wait 90 second between each set, it's really annoying. I read a book usually, or browse HN, but it's a very "interrupted" thing.
Well as Wolfram has stated, we need to mix LLMs with some deterministic reasoning. Real knowledge based output that can be routed to for these purposes. The hard part is determining when.
We need to go back to using the AI tech stack as a tool for (broadly speaking) automated pattern recognition, i.e. it can be great at image analysis and spotting early cancerous changes in lungs, but it ought to stop at pre-qualification stage and leave it up to the human to decide what the change is and what to do with it. What the AI crowd today want is all the photos in the world to be used to train models to recognise cancerous changes and then accept blindly the following diagnosis: "it is a cancerous change of human lung tissue, therefore the patient is a labrador with chickenpox and we recommend immediate removal of the patients left wing." When you tell them it's pure garbage, they come back with answers worthy of the slimiest orators in history and solutions that Rube Goldberg would not have thought of.
Why do people think inserting an LLM into the mix will make it better than just an evolutionary or reinforcement model applied? Who cares if you can talk to it like a human?
Yeah, when the author was writing about that initial query about delay-per-unit-length, I'm thinking: "This doesn't tell us whether an LLM can apply the concepts, only whether relevant text was included in its training data."
It's a distinction I fear many people will have trouble keeping in-mind, faced with the misleading eloquence of LLM output.
I think you are looking at the term generalizing and memorisation.
It have been shown that LLM generalize, what is important to know is if they generalized it or memorized it.
imo, it's the same reason that Grace Hopper designed COBOL to write programs instead of math notation.
What natural language processing does is just make a much smarter (and dumber, in many ways) parser that can make an attempt to infer the intent, as well as be instructed how to recover from mistakes.
Personally I'm a skeptic since I've seen some hilariously bad hallucinations in generated code (and unlike a human engineer who will say "idk but I think this might work" instead of "yessir this is the solution!"). If you have to double check every output manually it's not that much better than learning yourself. However, at least with programming tasks, LLMs are fantastic at giving wrong answers with the right vocabulary - which makes it possible to check and find a solution through authoritative sources and references instead of blindly analyzing a problem or paying a human a lot of money to tell you the answer to your query.
For example, I don't use LLMs to give me answers. I use them to help explore a design space, particularly by giving me the vocabulary to ask better questions. And that's the real value of a conversational model today.
I think you've nailed a subtly — and a major doubt — I've been been trying to articulate about code helpers from LLMs from day one: the difficulty in programming is reducing a natural language problem to (essentially) a proof. I suspect LLM's are great at transferring style between two sentences, but I don't think that's the same as proof generation! I know work is being done I this area, but the results I've seen have been weird. Maybe transferring style won't work for math as easily as it does for spoken language.
It can be scary to pigeon hole yourself into any specialist category with which it gives you less freedom of choice in the market place and be forced to make hard decisions on where you live and work. At least for me it's why I continually push towards being a generalist. That is, of course unless your talents in some specialist area are so fundamental to most companies that you'll be sought after for those skills and paid handsomely for the fact.
I disagree. Leetcode is less representative of your job than code review on a PR. Code reviews are an every day activity that can easily cause outages.
What's your argument that code reviews aren't a good representation of skills?
Reciting all x86-64 opcodes and their possible encodings from memory is even more difficult, so it must be an even stronger signal. Why not test programmers solely on this essential knowledge? It’s almost unavoidable that their code will execute on an Intel/AMD CPU at some point, so it’s best to filter for programmers who have the strongest grasp of the fundamentals.
If you let people know that they can get a $300k job by memorizing x86-64 opcodes, you will start getting >0 pass rates on this test. It’s a given.
When people have time to prepare for the test and the incentives are high enough, you only get what you measure. The leetcode fans don’t seem to understand this.
As someone who hires a ton of engineers, I wholeheartedly disagree with this.
I actually really like this concept. The main benefit is that this approach allows a more conversational interview and also allows to cover a broader range of real-world problems.
There is almost 0 correlation between candidates ability to be effective collaborator and their ability to solve leet code problems. The former is a lot more important to us than your ability to chew out code.
More than that: a lot of candidates who are good at leetcode challenges are good at them because they specifically practice them.
Does this translate to creative problem solving and thinking? To productivity when encountering novel tasks? To being a good teammate? There is a stronger correlation with "having free time" and "solving leetcode" than being a great engineering teammate, IMO.
I personally consider the ArgoCD UI an anti-feature. Attaching some hulking mass of Javascript dependencies to the thing that has cluster-admin rights to my production cluster is unnecessary attack surface for me.
ArgoCD also has its own auth system and permissions. You give ArgoCD cluster-admin rights, then it uses impersonation to pretend like it has lower permissions. One little bug there and you can trick ArgoCD into escalating your permissions, which happens a lot: https://github.com/argoproj/argo-cd/security/advisories/GHSA...
While not officially supported, you can technically deploy Flux with limited permissions, but ArgoCD's dependence on impersonation means it cannot run with lower permissions.
Redis is also a requirement to run Argo CD. When comparing load on my home server, flux was much lighter.
Flux also has a pretty cool terraform controller too.
At work though, we use Argo and our developers use its gui to get an overview on their applications.
