npalli

This idiotic churn used to be restricted to front end devs and backend folks were spared. Thanks for Agents now every part of the stack is going to be in constant churn thanks to resume builders. At no point in the article did you get a sense they have a handle on what their problem is and why a certain technology and architecture would be a better fit. Just try it out, churn out, try it out.

This is a strange release. The M6 tops out at 32GB RAM and then you fall back to M5/Pro. Kinda midway between an Air and a true Pro. Only 14" no 16" with M6.

Strange position in 2026. If there is any writing in the wall it is this. Nobody is going to be writing Rust code from scratch given where LLMs are already now, much less in 5 years. A golden rule of software is fresh coding is a small part of the lifecycle most of it is maintenance. You may be able to get people to write Rust code (vibe code it), however a tiny minority of them will want to maintain LLM generated Rust code!! Rust will most likely be an intermediate representation - a security/formality check of sort - before converted into the appropriate heterogenous code across GPU/CPU/TPU/XPUs. Even there some other better representation will supplant it as Rust's ideas will be found too limiting. The idea that you as a programmer will sit and write Rust source code is just going to look very archaic. Thank god for LLMs for saving everyone just in time.

Far from it. Until Rust can read C++ code seamlessly, there will be more code written in C++ than Rust for the foreseeable future. HPC, Finance, AI, Games, Databases, Hardware all continue to be written in C++.

Like all software, vibe coding is not the hard part. Maintaining it is. I'm sure there there 100 times the number of people wanting to vibe code this in Rust vs wanting to maintain someone else's vibe coded Rust. So, usually the natural state is keep running the show as-is until it is the next guy's problem.

Interesting, Amazon doing the same for AWS Trainium.

https://github.com/aws-neuron/torch-neuronx

Agree, I feel you only hear from people who hate Windows and for every such loud opinion, there are probably 20 who are quite satisfied but never mention it.

If no one is reading or writing Rust code at that point, it all being agent driven, Rust itself is going to be a very short step before it gets disinter-mediated away and it's English -> complex heterogenous machine code across GPU/TPU/CPU/xPUs. Not sure Rust fans or C++ haters have thought this through.

Congrats, what were some learnings in doing the C++23 port? Good and Bad.

You are right, sometimes the simplest answer is the right now. Don't trust Meta. Life is going to be much simpler. Fool me once, shame on you. Fool me continuously for 20 years, shame on me.

Alas, it writes so much better than the average human that it's what everyone started using. Hence the utter familiarity and now contempt.

How does one know whether they actually learnt physics this way. Seems a lot of reading and structure, that's great. Not convinced about whether there is a change in understanding. Typical university/college classes have assignments, quizzes, exams etc.. don't see the equivalent.

I laughed at this "but the code complexity became too much for me, and I abandoned the project." even Terry Tao finds some code too complex to write.

Really bullish on LLMs expanding code development by a very large group of people who are really smart in some domain but could not get into 'coding'.

Rewrite in C is the new Rewrite in Rust.

So, and iPad with an attached keyboard.

If it is this easy, surely the trend is Rust output being an intermediate pass of the LLM super compiler. A security subset if you will (like other kinds of optimization), it will move from Rust specs to some deeper level of analysis and output the final executable. Some brave souls will read the intermediate Rust output (just like people used to read the assembler output from compilers) but the LLM super compiler will just translate a detailed English like spec into final executables.

Good. How does it compare to cuOPT which seems to be 10-5000x faster than existing solvers for Vehicle routing if you believe the marketing.

Not Rust because you won't find anyone to maintain code. Rust is only for rewriting existing code in other language not maintaining it. C++ perhaps because of number of people who already know it, but only for performance sensitive areas. Also, lot of training data. Other than that, usual suspects, Java/C# for enterprise and node/Python for web/DS.

LOL, the blog gives a lot of detailed reasons, even summarizes it [1] and but some random stranger gives an outdated opinion from the '90s, which is not even wrong just plain humorous. If slave labor, how come everything else is also not so cheap.

   [1] Virtually all the major mechanisms that can drive efficiency improvements — improving technology and overlapping S-curves, economies of scale (including geometric scaling effects), eliminating process steps, reducing variability and improving yield, advancing towards continuous process manufacturing — are on display here

Andy is probably the only person who adores Larry Ellison (Oracle) unironically.

Seems like a bug, it's pulling in all the jobs that have either vision or pro in their text (which is probably everything). Need to put "vision pro" to filter and even select the product (Apple Vision Pro) to get about 74 jobs.

This is a good resource, however for about 99.99% of people, you are most likely to just use a foundation model like ChatGPT, Claude, Gemini etc. so this knowledge/training will get you neither here or there. I would suggest you look into another Karpathy's video -- Deep Dive into LLMs like ChatGPT.

https://www.youtube.com/watch?v=7xTGNNLPyMI

LOL, this is the list to keep in your head for this so called "manual". Best of luck of those who will work through this. BTW, Karpathy made that comment in 2025 not 2024.

  Morphability - natural language as morphable code
  Abstraction - tasks become reusable commands
  Recursion - stack abstractions for leverage
  Internal Consistency - prevent system drift
  Reproducibility - crash-resilient design
  Morphic Complexity - recognize over-engineering
  E2E Autonomy - measure actual capabilities
  Token Efficiency - maximize work per token
  Mutation & Exploration - controlled self-improvement

Seems you are paying the Dell tax of 15%. The same setup is $4K from NVidia, Lenovo and $3K for 1TB at Asus.

https://www.dell.com/en-us/shop/desktop-computers/dell-pro-m...

You just need to master one package managed in depth and you will get what you really want with Modern C++.

Just like San Francisco and Dallas/Texas (from his article) are very different in the US, we should expect lot of differences in Europe (as others mentioned, he clubs UK with EU). Housing is a general problem for all major cities though, not sure why you think it is unique to London in the whole continent. Stockholm, Paris, Dublin, Lisbon to name a few, are pretty bad for housing in their own unique ways. Certainly shouldn't be "breaking your brain".

Great summary of the year in LLMs. Is there a predictions (for 2026) blogpost as well?

Seems very detailed and comprehensive. Did I miss it, but was there a performance comparison to the PyTorch version at the top?