dangoodmanUT

Because huggingface is not charging them?

CFAA doesn't just mean the feds kick down your door, you actually have to get reported and sued over it.

throws away spec doc for my own

Are we all having this idea?

25k tokens is a few turns

J-space sounds oddly similar to...

finally, this was needed.

A big worry was always "why does workers sit in front of my cache? that's a waste of an invocation if i'm returning a cached result"

i believe the author is confused

this is the wrong direction

(jk i read the readme)

If you actually bench it locally (local S3, actually writes to disk for "staged" operations), ZeroFS performs horrifically. Ceph blows it out of the water. I don't have the exact numbers, but when I was building a toy CoW distributed block device and filesystem I did a perf matrix, and ZeroFS (even with an hour of codex tuning it) was never within the same 1 or 2 orders of magnitude perf-wise.

Sure, Ceph isn't "S3-backed", but when you're talking about an actual filesystem or block device (the thing that does lots of small-IO), you care more about io-perf than sequential.

Large sequential jobs that everyone is targeting now (ie AI workloads) can use s3 directly just fine, because they don't have decades of code built on top of the filesystem.

How are they going to be competitive with top models at 70B size?

iroh is consistently one of the most delightful projects i've ever worked with. The people reflect that too.

Congrats iroh team!

These are actually the most durable businesses because they have network effects: restaurants alright signed up, sellers onboarded, etc.

Easier for them to adopt AI than AI companies to rebuild the networks

Not comparing to GPT Pro models is a bit strange, considering that's the natural comparison

I'm really not looking forward to Gemini models on my devices.

Gemini models clearly gaslight the user and hallucinate, they're also SUPER verbose, as shown in the demos from the keynote.

Plus, if they're not charging a subscription for this, you know we're getting the dumbest models...

The Messages API now accepts system entries inside the messages array. Developers can update Claude’s instructions mid-task without breaking the prompt cache or routing the update through a user turn. This can be used in a given harness to update permissions, token budgets, or environment context as an agent runs.

Biggest deal imo

Exactly this

It has been 0 days since GCP has taken down a startup (again).

You see this at least once a year. Never heard of this from AWS or Azure.

In all seriousness, this is why we don't use them. They have the most ergonomic cloud of the big three, then absolutely murder it by having this kind of reputation.

This. Type casting is an insidious problem (both correctness, and perf)

Why pick Gemini? It's probably the worst tool calling model of the major labs.

Seems like AI slop. They list Rippling, and the description starts with Parker Conrad, but the rest of it is about Deel:

Rippling

Parker Conrad's redemption arc after Zenefits hit a plot twist when Rippling sued competitor Deel for planting an undercover spy inside Rippling who was paid €5,000/month by Deel's CEO to steal trade secrets. The DOJ opened a criminal investigation. Deel allegedly ran the same playbook at crypto HR startup Toku. YC uses Rippling for their own HR — awkward.

Per this description Rippling did nothing wrong here, all about Deel...

GCP would be perfect if they didn't have a history of randomly dropping quotas on startups, causing them downtime

If this goes through, it feels like it will stoke rust on zig violence

Two incidents? Just two?

In seriousness, looking at their scale, this is an insane engineering challenge.

Especially if they’re moving databases, not easy ever, and certainly not at that scale

shift+cmd click is the macos shortcut for "open this link a new tab and go to that tab"

Works on these psuedo links all over github

you can tell this was generated with Gemini, the way it loves to do those "enter on scroll" sentences

On this episode of "chatgpt just killed my startup"

Tune in for the next episode: Word

This feels like defi... de-ai

Yeah, it’s like drinking coffee when being really tired. You’re still tired, just “faster”, it’s a weird sensation.

Note that apples terms do not allow someone to sell something like an agent running on macOS. They have explicit cut outs for 24-hour minimum leases of full hardware, but they prohibit this with vms

This was inevitable, I called this a few weeks ago [1]. It’s an easy way to increase revenue without making the models smarter, and lock you in harder

https://danthegoodman.substack.com/p/where-agents-converge

This is marketed as competition. But it’s not. It’s a cartel. Each company gets its own protected territory, and consumers get no choice.

Starlink has me extra excited for exactly this: region agnostic competition.

Get rekt, telecom incumbents

Don’t try to sell developers the stuff that’s fun, the stuff they want to build. Sell them the stuff they avoid or don’t want to take responsibility for.

I think a lot of people in devtools could learn from this right now