Dave Guarino

Working to safely deploy AI in state & local government at Anthropic. Ex: CfA/GetCalFresh, CA, USDOL, Propel. Toddler dad; ferry maximalist. Posts solely mine.

[Dave Guarino @allafarce]

Okay, I've *finally* sent out a newsletter.

I'm trying out Substack. Let's see if it works.

In this one:
- Personal updates
- A theory of 3 constraints in tech services
- Medi-Cal friction mapping

https://t.co/PiitYQHWRa

[Dave Guarino @allafarce]

My daughter loves Oingo Boingo (on vinyl)

I have succeeded as a father.

[Dave Guarino @allafarce]

@alexolegimas I believe Zachary’s in Berkeley does this at lunch!

[Logan Graham @logangraham]

Within ~2 years, there might be >0.5 Manhattan Projects worth of philanthropic $$$ to spend on the biggest challenges, like cyber or biodefense.

@nanransohoff is right: the hard part is finding enough good organizations to spend the $.

You should build one!

[Dave Guarino @allafarce]

@TheStalwart Very cool to see you testing models like this! (In general I wish lots more people tested models, it’s useful and interesting and teaches you a lot.)

[Dave Guarino @allafarce]

We are constrained first by what we believe to be possible, and then, only subsequently, by what is in fact possible.

(And as the latter changes we really, really need to update the former quickly.)

[Andrew Curran @AndrewCurran_]

Head of Frontier Red-Team at Anthropic.
https://t.co/0vi45EJ903

[Derek Thompson @DKThomp]

Skepticism of corporate marketing and AI boosterism is always warranted, but I think the folks who accused Anthropic of overrating Mythos should check out this post by Mozilla developers indicating that the Firefox team fixed more security bugs in April using Mythos than in the past 15 months combined.
https://t.co/0hmpnz0pQZ

[Dave Guarino @allafarce]

@rohindhar I’d be curious to see duplexes (admittedly a more complicated category) if that data exists/is easy!

[Dave Guarino @allafarce]

@johnjhorton I would be interested to hear the longer form/mechanism-level theses of yours on this!

[Dave Guarino @allafarce]

@sebkrier Fun fact CS 1.5 fully runs in browser these days (I may or may not have fallen into this rabbit hole a few years ago)

[Dave Guarino @allafarce]

I’ve worked at Anthropic for three weeks and I can say it is both wonderful and *quite* different from any other place I’ve ever worked. Feeling immense gratitude to get to work on the things I do (and concomitant obligation/responsibility.)

[ClaudeDevs @ClaudeDevs]

Over the past month, some of you reported Claude Code's quality had slipped. We investigated, and published a post-mortem on the three issues we found.

All are fixed in v2.1.116+ and we’ve reset usage limits for all subscribers.

[Saffron Huang @saffronhuang]

Fast follow on my "What 81k people want from AI" work — this time, focused on the economic stuff: who's worried about their job, who's getting faster, and where the productivity gains are landing.

Findings include:
1. People are quite well-calibrated on personal job threat when we compare to observed job exposure
2. The biggest productivity gains go to the highest _and_ lowest-paid jobs

[Jeremy Howard @jeremyphoward]

Wow I can already say after just 5 hours using @AnthropicAI Opus 4.7 that this is the first model that "gets" what I'm doing when I'm working. It feels aligned with me in a way no previous model did.

(4.6 actively worked against me. I hated it. So this is *very* exciting!)

[Dave Guarino @allafarce]

(Oh also my first day was last Tuesday. Interesting first day!)

[Dave Guarino @allafarce]

Some news! Last week I joined @AnthropicAI to help state and local governments safely and effectively deploy AI capabilities.

Personally I see profound opportunity to transform how public services are delivered & accessed—and I feel a deep responsibility to help that go well.

[Dave Guarino @allafarce]

@geoffreylitt @AnthropicAI Thanks so much!

[Dave Guarino @allafarce]

@jackclarkSF @AnthropicAI Thank you, and likewise!

[Dave Guarino @allafarce]

@markasaurus @ryxcommar @michaelrbock Oh yeah, for sure Claude makes mistakes! But what's so useful to know here is if it requires additional context (eg docs, web search), steering (plugin), or if intelligence/reasoning gets it there. Different paths, and this is an analogous use case to many of interest to me!

[Dave Guarino @allafarce]

@ryxcommar @markasaurus @michaelrbock Thanks! If you try Opus on same query/task id be really curious to see if the intelligence difference changes it