HN user
Read through it an I'm curious whether setting the date and cmd on every system prompt call will cause the cache to invalidate.
I guess the cache would only be invalid if the day changed or the root directory, which would technically happen infrequently enough.
Isn't Apple about to license some variation of this from google for on-device AI? Maybe it’s their sales pitch to Apple and then they will lock it down.
Maybe it's my experience, but TPMs were often responsible for coordinating large org-wide or cross-org initiatives. It's prohibitively expensive to have TPMs on anything smaller.
Since engineering with AI is still very technical, I would wager that software engineers would stretch into less technical areas of software development rather than TPMs stretching into technical areas. I only say this as someone with experience with AI and I see how easy it is to write bad code with AI if you're not aware of what it's doing.
Engineers are not going anywhere, they are going to fill the spaces outside of coding that are still critical to shipping new product.
Here is a post that summarizes what I mean: https://substack.com/home/post/p-200064883
Yes, size and performance are not only problems for local LLMs, they are problems for frontier LLM companies like OpenAI and Anthropic. The latter still lose a ton of money on inference and advances in efficient, performant models helps their bottom line.
They are addressed but the core of the thesis is still wrong:
This is the core problem: our entire evaluation infrastructure is structurally reactive. We measure the system after it has changed. We never predict the change.
That's kind of the point of evals.
The argument in the article is backwards. Evals test the stability and boundaries of a concept. They are not created before the concept has been prototyped (which the author acknowledges).
An eval is not somehow breaking silently due to some new capabilities in an LLM. It wouldn't be a good eval if it did. What it does is steer the LLM towards specific goals. If anything, an argument can be made that they restrict creativity and experimentation by narrowing goals.
If the argument is that evals need to written before some new behavior can be devised, that's incorrect. There are an infinite number of evals that test for things which cannot be done. Only when something has been demonstrated to work in a specific context, can an eval be written.
How does this compare to using Claude Web with connectors to build the same feature?
On a separate note, READMEs written by AI are unpleasant to read. It would be great if they were written by a human for humans.
I applaud the move. It's also a little disingenuous to talk about moral standings when the third opening sentence is "The math hasn’t worked out for a while now." If the numbers were working out, would they continue to turn a blind eye on the privacy tracking?
Too much of code is data transformation. input -> sanitation -> db -> consumer -> api -> client. Business logic defines the shape of that data and some service-level rules but the majority is just shoveling data.
Those are raw numbers. I would look instead at the job changes over total employment numbers. I don't have the numbers but I would wager we have many more people working in tech today (overall) than we did in 2008.
Also, that spike in 21/22 really did a number on people's expectations. The one constant in this industry is its cyclical nature.
Used, yeah
This laptop competes against M2/M3 MacBook Airs. Going to be hard to justify a Neo when the others are so much more powerful.
Wishing the best for all those affected and excited to see many of you start new companies and continue to innovate.
Why anyone is still using X after 2025 is a mystery (I know, it's where everyone is, but the moral implications are wild)
Publishing is more than just authoring. You have research, drafts, edits, source verification, voice, formatting, multiple edits for different platforms and mediums. Each one of those steps could be done by AI. It's not a single-shot process.
Don't let common sense stop you from a good time.
We’re about to see if LLM regress or evolve
That's right, I said TypeScript but yeah, it's v8 under the hood.
Not quite systems programming but this might give you some insight. Swift is memory efficient, and runs stable backend services. I've seen benchmarks showing that it's slightly more performant than typescript but twice as memory efficient (but not as efficient when it comes to memory management compared to Rust, C, and C++).
The other point I've seen is that its string library is slow and very accurate.
Besides that, the C-interop means you have quite a bit of flexibility in leveraging existing libraries.
Nothing particularly insightful other than avoiding messing with previous messages so as not to mess with the cache.
I asked Claude to summarize the article and it was blocked haha. Fortunately, I have the Claude plugin in chrome installed and it used the plugin to read the contents of the page.
Before people in this country can unionize, they have to see their peers and fellows, not as strangers. The structures in this country have put each person against the other in pursuit of wealth.
Engineers are happy to pay for tools (hello, Claude Code). Libraries are quite different and it's a little uncomfortable to build a business on a closed-source, proprietary library.
This is amazing! Thank you for open sourcing the project. It must have been a hard decision.
Would be great if you didn't need to believe in a supernatural being.
The US is structured to promote loneliness.
If you want to fix it:
- More free public spaces (parks with benches, squares)
- More free public events and activities (free concerts, art installations, plays)
- Greater physical proximity (it's hard to make eye contact if everyone drives)
- Wealth distribution (create a society where one's value is not based on their net worth)
- Encourage days off for community service
In other words, provide socially-funded incentives for people to be close to one another physically and remove income as a measure of value.
A one-time wealth tax doesn't seem to make a ton of sense.
The current loop hole is that loans are not taxed so people take out loans against their assets thus never paying income tax. Tax those loans and refund taxes when the loans are repaid.
An article mentioned that it will be an iOS chip
The M5 Pro/Max models are likely going to arrive in March (but maybe earlier)