nijave

I don't think so. I have lieer + notmuch setup with a few AI generated Python scripts that will mark things as read. It also took about 5 minutes of fiddling with notmuch cli and jq to extract SNS message bodies from emails (for instance).

I think a thin layer on notmuch would get you pretty close without so much infrastructure.

On the web scraping side, scrapling seems to pretty much just work. There hasn't been a website yet it won't load in some mode--even Google SERP.

I think people are making skills to have AI generate text that sounds like them but AI--especially Claude--has a hard time not sounding like AI.

Even if latency is improved, it's still a monumental task powering a latency sensitive safety critical system over the internet--especially one that's moving.

Maybe if latency can be improved _and_ it can run local inside the vehicle.

>could imply curtailing the model's creativity and 'thinking out of the box' capacity

Imo a feature. I can handle the creativity and out-of-box thinking just fine. I want the model to do the grunt work but be smart enough to do the grunt work correctly.

In fairness (in my experience) Fable actually generates useful output while incinerating tokens.

Sent Opus 5.5 an example that used go context.WithTimeout and it tried to tell me that was wrong and I should pass timeouts as ints before finally admitting the docs it cited didn't say to use ints and that's a ridiculous design in go anyway (it was trying to claim that was codebase convention--passing ints...)

The old analysis paralysis

It's been like this at least months. I turned telemetry off a few months ago and it basically prevented all stages rollouts of new features from working.

Anthropic commonly gates behind feature flags that require telemetry until they're "promoted" and defaulted on.

A little bit annoying you can't manually control the flags without telemetry but I think the title is a bit click bait.

Luckily all the other LLM providers are also still making progress with less onerous "safeguards"

>I haven't managed to get one for level "max" yet, it hit the limit of 128,000 cap for output tokens while it was still reasoning about the question!

Off to a _great_ start...

Also interesting this somewhat mirrors my recent experience with Opus 5--too much effort and it starts looking for things to do and invents requirements that never existed

Didn't these used to be called Argo Tunnels a long time ago? I think Argo Tunnels got repositioned for connecting to origin servers without needing an incoming connection to the origin but I thought they originally did something like this/ngrok

Yeah, I just checked the page >can connect their server to the Internet with Argo Tunnel for free

Don't think we had any issues with Warp but Cloudflare APIs are a bit of a mess. Seems like they've cobbled together and then renamed/rearranged things multiple times now and it's a bit hard to keep up.

Terraform has worked decently well especially since there's a few random settings here and there that aren't exposed anywhere in the UI (facepalm)

Yeah but you lose anonymity unless you can devise a scheme to identify advertisers and bots separately from and without identifying real humans.

Last I checked, they just blanket added `cache-control: max-age=0` to everything.

Loading https://gitlab.com/gitlab-org/gitlab is showing 14 API calls so presumably significantly more DB queries for a public page

RIP NAT

https://gitlab.com/gitlab-org/gitlab

Makes 12 graphql API calls and 2 /api/v4 calls... So you get like 4 pages per hour unauth?

Presumably but I wish they'd also focus on optimizing code/making pages fully cacheable instead of rate-limits and blocking

I can pay Amazon pennies on the dollar to saturate 100Gbps with arbitrarily high concurrency

Sometimes I also want a smaller version for when I don't need that performance but don't want multiple implementations of storage backends

Edit: before someone says NVMe, highly available across 3 fault domains

For testing/dev setups you can turn sync off and get some performance back but that's only useful for cases where losing data is okay.

I think it's possible to get an LLM to run your business but I think it's still prohibitively expensive to spend the time and money getting guardrails, skills, knowledge, context setup.

Getting it to work reliably is quite the feat. AI coding is a much lower bar when there's (ideally) still a human expert that can set it straight when it goes awry.

I also think some of these debates get hung up on what's possible and ignore what's feasible and practical.

I usually use Fable with superpowers to generate plans and do subagent driven development but unfortunately I was at my Fable weekly limit with some other usage left.

I forgot how awful using Opus 5 directly is.

Knowing Meta, I'd be more surprised if they _didn't_ distill frontier models than if they did...

Not a lawyer but distillation sounds like a transformative work.

Same thing as Cliff Notes imo. In every other area of manufacturering and tech I can use a machine to build a new machine that competes with the original machine. Should Milwaukee be able to prevent DeWalt from using their drill to make a competing drill? Should Jetbrains ban Eclipse contributors from using their IDE?

Just earlier Opus 5 was supposed to migrate a Python codebase from kafka-python client-per-send to a client singleton. It decided it should add a partition key "that nobody asked for" which it admitted to at the end of the implementation.

It randomly decided going from no explicit partition key which would randomly add to a Kafka partition to partioning key company was the right solution.

In fact, that was fairly counterproductive since this codebase dumps events to the topic per company so you'd get a lumpy distribution where only one partition is ever active at once (starving other consumers)

The fact it decided it should add this without prompting was a bit shocking and I'll probably try to make it explicitly clear in the context not to "do random shit that wasn't asked for"

Probably need a lawyer to weigh in but "originate from and be understood by a contributor" sounds like "write a prompt". The machine isn't sentient and doesn't just decide to generate code one day.

Maybe that's not what the authors of the policy intended but that's how it reads imo

Enterprise contracts

Providers can compete on cost which is a small but positive outcome. It also encourages providers to find more efficient ways to serve models which also lowers cost.

That said, these frontier models are still astronomically huge.

State sponsored maybe but it looks to take at least 6Ti of VRAM to run which is quite costly...

Seems fine to me. Maintainer wants to use AI and community bans many AI uses so maintainer leaves.

As AI bans go, this one doesn't seem too stringent if you disclose AI and actually check what it's doing.

Seems like the most practical path is some kind of distributed peer to peer contraption where you can allocate some storage and optionally participate in crawling.

I imagine if set some constraints you could get index size down quite a bit but I still suspect it'd be hard keeping up with content churn.

Edit: Looks like enwiki bz2 is coming in around 46Gi which isn't too bad considering the amount of content it contains.

SearXNG has been working pretty well for me. I had an agent write the MCP then do a couple passes comparing to server side LLM web tools and exa and tweaking and it works pretty well. I also added scrapling for fetch which covers pretty much everything but sometimes is a bit context heavy.

Most of the stuff should be bypassable with browser automation but you'd need more compute to run a full browser versus a basic uri fetch.

I imagine it'd take quite a bit of shape for the index and be hard to keep it up to date unless you restrict what it indexes.