curious_curios

I love the effort to help modernize go code! I feel like it might have been better if they leveraged the Analyzer framework to get this stuff into `go vet` and `go fix`, which would be infinitely easier for LLMs and developers alike to use.

Alternative take: I have not met a serious craftsperson software engineer who did not deeply care about some KPIs. They might not have called them KPIs, but there were numbers they tracked constantly.

In the example of being asked what testing KPIs to tracked, why wouldn’t they include:

- Flake rate - Test run time, a large component of CI lag

Improv believe it or not.

Though with gas, maintenance, insurance on top of the car payment, that $700 is probably the floor of car ownership.

Many of the open weight models are trained on outputs from these models (distillation)

I would love for this book to be updated with all the work on adding generative AI to the software engineering workflow at Google.

The Enchiridion is perhaps the most distilled and practical manual of practical stoic philosophy and one I wish I had discovered earlier in life.

While cool sounding, the project is two weeks old and contains lots of made up terms. It’s safe to say it’s a figment of an AI’s imagination.

This is pretty standard for specification documents, probably more accurate to say AI sounds like them than the other way around.

Ignoring the particular technologies used (OAuth/JWT) it looks like they’re adding more auth to the devices themselves; think two computers connected to the same network switch not being able to impersonate each other.

Reading between the lines I suspect the following:

- We’ve already demonstrated breaking 2048-bit RSA in tightly controlled lab environments

- It’s expected that will be possible at scale by 2029

- The US expects to have around 2 years of supremacy in the space before adversaries catch up

Not to be confused with Pharo, the immersive programming experience: https://pharo.org/

Generally they do it via bitemporal modeling in databases: https://en.wikipedia.org/wiki/Bitemporal_modeling

“There are only two kinds of languages: the ones people complain about and the ones nobody uses.”

- Bjarne Stroustrup

Whoever wants to write the underlying engines for virtually every browser: Apple and Google. They both have their agendas that they try to push via them.

If their experience mirrors my own there will be a follow up post, “Why having 50 individuals on your cap table is a pain, I should have gotten them into a syndicate”

Depends on how you define big, but there’s Gemma, Phi, OLMO, Mistral and GPT-OSS that are all competitive and can run on commodity hardware.

A lot of it comes down to the context and a prompting strategy tailored to the particular model. I don’t believe the current benchmarks really take those optimizations into account.

I’ve personally been getting better results with Gemini as well, but I think it’s just because I’ve used it more.

I at least respect them for reporting them. It feels like lots of cloud providers don’t, or begrudgingly.

I've actually been getting a lot of mileage out of textproto: https://protobuf.dev/reference/protobuf/textformat-spec/

Well typed, simple syntax. Maps are annoying though.

Given that AWS, Azure and GCP are all recording 20-40% YoY growth, no, I don’t think they’re losing ground.

As for startup credits, they’re still handing out $100-200k like candy if they deem you a serious startup. There was a lot of abuse in the past so they started putting up filters.

When you fly in an airplane do you care what engines they’re using? Do you care what avionics subsystems are installed and what SoCs they utilize?

What makes this situation any different? It’s just that LLMs are the new hotness. That will fade.

Previous discussion: https://news.ycombinator.com/item?id=45798827

Mail/calendar/chat

If you have a moderately successful app, sdk or browser extension you will get hit up to add things to it like this. I think most free VPN services also lease out your bandwidth to make their money as well.

- Get into a reputable accelerator program, you’ll build your network.

- Find meet ups in your major metropolitan area.

- Your 2nd and 3rd degree networks are bigger than you imagine, you can absolutely get a warm intro.

- Forgo all investment and bootstrap.

If you’re getting value from it, keep adding it!

Two cases where this advice breaks down:

- If the telemetry boilerplate code starts hampering your ability to read and reason about the underlying business logic code.

- When your observability bill gets too big.

I think you meant the Go 1.25 release notes?

- Ads

- Paywalls

- Subscribe pop-ups

- Obvious AI slop

- Dead links

- Information already out of date