eclectic29

Yes, but you expect much better from a seasoned engineer.

One man’s extremely horrible mistake and look how the whole industry has been suffering for decades - countless articles debating the existence or absence of comments, new formats, new parsers, new ways of adding comments, post-processing, etc. What a pity!

I just installed AdGuard and the ads didn’t go away. What should I be uisng? I have an iPhone.

If Martin Kleppmann is the author I know this stuff will be worth watching out for.

I mean if it doesn't fit the team's or the project's stage, don't do them daily. Simple! I just solved it for you. Not sure why we need tons of articles on scrum to state the obvious :-). Are the engineering teams so dumb that they're following scrum to the letter? Come on!

Slightly off topic: Learnt a new word today 'vulgarization' which seems to have a completely different meaning from the obvious. Thanks.

Hahahaha, what a joke Google has become! Another half baked name change. Looks like no one really cares anymore at Google.

This is excellent. Thanks for sharing. It's always good to go back to the fundamentals. There's another resource that is also quite good: https://jaykmody.com/blog/gpt-from-scratch/

These days, the roles I consider are in leadership so if we lack vision and a clear understanding of our value I’m usually empowered to fix that. If you’re interviewing for a more IC role, your hiring manager and teammates being unable to communicate expectations and success criteria is obviously a bigger concern.

So, the author doesn't seem to consider IC role a leadership role. I see, ok.

A whole article about getting a job in tech and no mention of the abominable leetcode. Surprise! From junior to principal Leetcode is here to stay. Good luck solving 2 problems in 35 mins.

Meta is less leetcode oriented? I'm shocked to read this. Meta is the poster child for leetcode style interviews. Meta requires you to solve 2 leetcode style questions in 35 mins (out of 45 mins - first 5 mins for initial pleasantries, last 5 mins for asking questions). For each question, you're required to (based on the signals they look for) ask clarifying questions, present a solution to the interviewer, get buy-in, code, verify with test cases - all this in 17.5 mins/question. Go figure! :-)

Even though the title has the word "translation" in it my mind ran to Apache Kafka :-).

Glad that people are realizing the ChatGPT/AI hype that is driving all companies nuts by creating unnecessary peer pressure to integrate some kind of GenAI in their products even if it doesn’t make sense.

Why are Meta quarterly results so special that they always enjoy the front page of HN?

Thanks for sharing. This is so disheartening. The other day I was debating with my friends how a doctors life is so cool that they don’t have to go though grueling coding interviews every single time they want to change jobs, don’t have to prove themselves every single quarter, don’t have to be answerable to anyone or write performance reviews or be subjected to arbitrary rubrics. Boy, I was so wrong. Every procession has its hazards. This has been a learning for me. Although I do feel that the PCP doctors in US seem to have a simpler life. They leave office at 5 and don’t take calls in the night. Happy to be corrected though.

Did you create a throwaway account specifically to write a snarky comment here? Looks like this practice is becoming more common.

Automatic resource management thanks to RAII. To me this is the most important thing.

Not yet.

Thanks!

It’s 100% truth my friend. Denying or flagging my comment doesn’t help anyone.

Horrible article. So many things are wrong. As with any thing in this industry, there's a pretty big spectrum and nothing can be generalized. Someone with a decade of experience simply cannot write such an article. This is click bait at best.

This is really bad news for Microsoft culture. See https://archive.is/ZuvYu. Also from Wikipedia: In August 2019, Suleyman was placed on administrative leave following allegations of bullying employees.[17][18] The company hired an external lawyer to investigate, and shortly thereafter Suleyman left to take a VP role at parent company Google. An email circulated by DeepMind's leadership to staff after the story broke, as well as additional details published by Business Insider, said Suleyman's "management style fell short" of expected standards.[19][20].

Try writing high performance math in Java even for CPU.

I acknowledge C++'s safety concerns, but no, Java is definitely NOT running rings around C++ perf. Not a single AI/ML model is implemented in Java. The core of AI/ML runs on C++ only. You may see a lot of Python, but the core engine that Python is wrapping is written in C++. Sorry to break it to you but Java cannot even come close here.

With all due respect Stroustrup seems completely delusional about real software out there in the wild. He needs a dose of reality.

Not angry. Having a discussion :-). It just amazes me how the HN crowd is more than happy with just trying out a model on their machine and calling it a day and not seeing the real picture ahead. Let ignore perf concerns for a moment. Let's say I want to run it on a shared server in the enterprise network so that any application can make use of it. Each application might want to use a model of their choosing. Ollama will unload/load/unload models as each new request arrives. Not sure if folks here are realizing this :-)

And what will you do after trying it? Sure, you saved a few mins in trying out a model or models. What next?

Of course, you can support concurrent requests. But Ollama doesn't support it and it's not meant for this purpose and that's perfectly ok. That's not the point though. For fast/perf scenarios, you're better off with vllm.

FWIW Ollama has no concurrency support even though llama.cpp's server component (the thing that Ollama actually uses) supports it. Besides, you can't have more than 1 model running. Unloading and loading models is not free. Again, there's a lot more and really much of the real optimization work is not in Ollama; it's in llama.cpp which is completely ignored in this equation.

Few hundred on the planet? Are you kidding me? We're asking enterprises to run LLMs on-premise (I'm intentionally discounting the cloud scenario where the traffic rates are much higher). That's way more than a hundred and sorry to break it to you that Ollama is just not going to cut it.