Dave Guarino
@allafarce.bsky.social
Software and complex (not complicated) systems. Pursuing the public good, sometimes with technology.
Google Gemini 3 Pro is out today. I ran through my standard "hard SNAP question" which I use to eval new models and it... absolutely excels.
(If you've got a hard public policy/benefits question you want to test, leave a comment and I'll run it through as I have time.)
(If you've got a hard public policy/benefits question you want to test, leave a comment and I'll run it through as I have time.)
November 18, 2025 at 5:39 PM
Google Gemini 3 Pro is out today. I ran through my standard "hard SNAP question" which I use to eval new models and it... absolutely excels.
(If you've got a hard public policy/benefits question you want to test, leave a comment and I'll run it through as I have time.)
(If you've got a hard public policy/benefits question you want to test, leave a comment and I'll run it through as I have time.)
My company, Propel, is doing a direct cash transfer to some high need SNAP families given November deposits are indefinitely delayed due to the shutdown.
Propel is contributing $1 million in company funds and looking to raise another $9m from partners.
www.propel.app/shutdown-fun...
Propel is contributing $1 million in company funds and looking to raise another $9m from partners.
www.propel.app/shutdown-fun...
October 29, 2025 at 5:26 PM
My company, Propel, is doing a direct cash transfer to some high need SNAP families given November deposits are indefinitely delayed due to the shutdown.
Propel is contributing $1 million in company funds and looking to raise another $9m from partners.
www.propel.app/shutdown-fun...
Propel is contributing $1 million in company funds and looking to raise another $9m from partners.
www.propel.app/shutdown-fun...
More progress on using Claude Code for my absolutely-personally-tailored daily todo list tool.
First bit was lots of keyboard shortcuts a la vim. Next I want a voice interface (via Apple Watch) so I've prototyped that with a text-based chat interface that will eventually be driven by voice.
First bit was lots of keyboard shortcuts a la vim. Next I want a voice interface (via Apple Watch) so I've prototyped that with a text-based chat interface that will eventually be driven by voice.
September 13, 2025 at 3:53 AM
More progress on using Claude Code for my absolutely-personally-tailored daily todo list tool.
First bit was lots of keyboard shortcuts a la vim. Next I want a voice interface (via Apple Watch) so I've prototyped that with a text-based chat interface that will eventually be driven by voice.
First bit was lots of keyboard shortcuts a la vim. Next I want a voice interface (via Apple Watch) so I've prototyped that with a text-based chat interface that will eventually be driven by voice.
I'm increasingly convinced Claude Code is an early glimpse of what general purpose AI with agency feels like.
I'm experimenting with making it usable for non-programmers:
- Created a custom output style called "non-technical user"
- It explains what it's doing in lay terms
I'm experimenting with making it usable for non-programmers:
- Created a custom output style called "non-technical user"
- It explains what it's doing in lay terms
September 3, 2025 at 11:31 PM
I'm increasingly convinced Claude Code is an early glimpse of what general purpose AI with agency feels like.
I'm experimenting with making it usable for non-programmers:
- Created a custom output style called "non-technical user"
- It explains what it's doing in lay terms
I'm experimenting with making it usable for non-programmers:
- Created a custom output style called "non-technical user"
- It explains what it's doing in lay terms
I've been using a pattern I call "Claude Compadre" — using Claude Code, but for decidedly non-coding tasks.
This approach means I can have Claude quickly "grow" its own capabilities via code generation.
Example here: generating code to hit the Asana API to help me with family task planning
This approach means I can have Claude quickly "grow" its own capabilities via code generation.
Example here: generating code to hit the Asana API to help me with family task planning
August 31, 2025 at 10:42 PM
I've been using a pattern I call "Claude Compadre" — using Claude Code, but for decidedly non-coding tasks.
This approach means I can have Claude quickly "grow" its own capabilities via code generation.
Example here: generating code to hit the Asana API to help me with family task planning
This approach means I can have Claude quickly "grow" its own capabilities via code generation.
Example here: generating code to hit the Asana API to help me with family task planning
It is time. Hello, Claude, my friend.
August 30, 2025 at 7:39 PM
It is time. Hello, Claude, my friend.
Using Claude Code to check logs for suspicious activity
August 25, 2025 at 9:55 AM
Using Claude Code to check logs for suspicious activity
Excellent bumper sticker.
July 17, 2025 at 2:52 AM
Excellent bumper sticker.
In my newsletter today I sketched out my own AI moonshot, a hyper niche one:
Can we use today’s AI coding capabilities not for greenfield app code generation, but rather for automated characterization test coverage on large legacy systems we’re afraid of changing?
Can we use today’s AI coding capabilities not for greenfield app code generation, but rather for automated characterization test coverage on large legacy systems we’re afraid of changing?
July 15, 2025 at 5:44 PM
In my newsletter today I sketched out my own AI moonshot, a hyper niche one:
Can we use today’s AI coding capabilities not for greenfield app code generation, but rather for automated characterization test coverage on large legacy systems we’re afraid of changing?
Can we use today’s AI coding capabilities not for greenfield app code generation, but rather for automated characterization test coverage on large legacy systems we’re afraid of changing?
A moment of appreciation for my morning walk.
July 14, 2025 at 3:20 PM
A moment of appreciation for my morning walk.
I need a political scientist to look into this, stat
July 12, 2025 at 3:26 AM
I need a political scientist to look into this, stat
You can just… learn to cook.
(Sautéed zucchini with salt, pepper, olive oil, lemon juice, chili flakes, mint, cilantro)
(Sautéed zucchini with salt, pepper, olive oil, lemon juice, chili flakes, mint, cilantro)
July 8, 2025 at 1:44 AM
You can just… learn to cook.
(Sautéed zucchini with salt, pepper, olive oil, lemon juice, chili flakes, mint, cilantro)
(Sautéed zucchini with salt, pepper, olive oil, lemon juice, chili flakes, mint, cilantro)
Underneath this sign lies a rich and hilarious story of a no good, very bad afternoon someone had.
June 12, 2025 at 10:04 PM
Underneath this sign lies a rich and hilarious story of a no good, very bad afternoon someone had.
Me when I pick up the Zuni Cafe cookbook and read that they salt their chicken TWO TO THREE DAYS ahead of cooking time:
April 27, 2025 at 9:53 PM
Me when I pick up the Zuni Cafe cookbook and read that they salt their chicken TWO TO THREE DAYS ahead of cooking time:
We evaluated a number of different models (OpenAI models, Anthropic Claude Sonnet versions, and 2 versions of Google Gemini) and saw some initial front running by the Gemini models.
(Note, this is very much an illustrative starting eval, not comprehensive!)
(3/n)
(Note, this is very much an illustrative starting eval, not comprehensive!)
(3/n)
April 23, 2025 at 10:29 PM
We evaluated a number of different models (OpenAI models, Anthropic Claude Sonnet versions, and 2 versions of Google Gemini) and saw some initial front running by the Gemini models.
(Note, this is very much an illustrative starting eval, not comprehensive!)
(3/n)
(Note, this is very much an illustrative starting eval, not comprehensive!)
(3/n)
Using AI to iteratively generate abstract representations of James C. Scott's "Seeing Like A State"; I rather like this one (DALLE)
April 22, 2025 at 3:30 AM
Using AI to iteratively generate abstract representations of James C. Scott's "Seeing Like A State"; I rather like this one (DALLE)
I don't post pictures of our daughter here, but I'll make an exception for this one.
(She is a literal bundle of joy.)
(She is a literal bundle of joy.)
March 26, 2025 at 11:02 PM
I don't post pictures of our daughter here, but I'll make an exception for this one.
(She is a literal bundle of joy.)
(She is a literal bundle of joy.)
Another lens on language models is that they are an incredibly capable ETL engine for semantics.
Here: transform a recipe into mise en place steps
Here: transform a recipe into mise en place steps
March 23, 2025 at 5:38 PM
Another lens on language models is that they are an incredibly capable ETL engine for semantics.
Here: transform a recipe into mise en place steps
Here: transform a recipe into mise en place steps
One of the less common situations where Claude failed me. Luckily Gemini worked like a charm.
Scan Costco receipt, grab the prices for select items, sum, divide by 2
(Unsurprisingly this is splitting our Costco run)
Scan Costco receipt, grab the prices for select items, sum, divide by 2
(Unsurprisingly this is splitting our Costco run)
March 2, 2025 at 7:25 PM
One of the less common situations where Claude failed me. Luckily Gemini worked like a charm.
Scan Costco receipt, grab the prices for select items, sum, divide by 2
(Unsurprisingly this is splitting our Costco run)
Scan Costco receipt, grab the prices for select items, sum, divide by 2
(Unsurprisingly this is splitting our Costco run)
Oh nice. I've been trying to go in that direction more, though I usually have farro standalone in a batch for the week then do a light garlic sautee w/ salt/pepper/EVOO/lemon juice.
FWIW this bean recipe changed my bean game entirely. Far better than standard approach IMO.
FWIW this bean recipe changed my bean game entirely. Far better than standard approach IMO.
March 2, 2025 at 4:02 PM
Oh nice. I've been trying to go in that direction more, though I usually have farro standalone in a batch for the week then do a light garlic sautee w/ salt/pepper/EVOO/lemon juice.
FWIW this bean recipe changed my bean game entirely. Far better than standard approach IMO.
FWIW this bean recipe changed my bean game entirely. Far better than standard approach IMO.
I appreciate Marina speaking very openly about what a lot of us have felt (absolutely including myself)
https://reason.com/2025/02/13/i-tried-to-fix-government-tech-for-years-im-fed-up/
https://reason.com/2025/02/13/i-tried-to-fix-government-tech-for-years-im-fed-up/
February 13, 2025 at 5:35 PM
I appreciate Marina speaking very openly about what a lot of us have felt (absolutely including myself)
https://reason.com/2025/02/13/i-tried-to-fix-government-tech-for-years-im-fed-up/
https://reason.com/2025/02/13/i-tried-to-fix-government-tech-for-years-im-fed-up/
Maybe I really should start a “Cooking with Claude” blog or something.
I’ve never browned butter (doing sage) and so I… just show it pictures and let it tell me if it’s browned yet.
I’ve never browned butter (doing sage) and so I… just show it pictures and let it tell me if it’s browned yet.
February 10, 2025 at 1:54 AM
Maybe I really should start a “Cooking with Claude” blog or something.
I’ve never browned butter (doing sage) and so I… just show it pictures and let it tell me if it’s browned yet.
I’ve never browned butter (doing sage) and so I… just show it pictures and let it tell me if it’s browned yet.
I earnestly believe one of the best ways to assess AI models is to use them to augment your cooking. Why?
- Low risk
- Can add value at almost any level of cooking ability
- If you know cooking well, you can see where it does well or falls down (and how specifically)
I use Claude daily for this.
- Low risk
- Can add value at almost any level of cooking ability
- If you know cooking well, you can see where it does well or falls down (and how specifically)
I use Claude daily for this.
February 1, 2025 at 10:04 PM
I earnestly believe one of the best ways to assess AI models is to use them to augment your cooking. Why?
- Low risk
- Can add value at almost any level of cooking ability
- If you know cooking well, you can see where it does well or falls down (and how specifically)
I use Claude daily for this.
- Low risk
- Can add value at almost any level of cooking ability
- If you know cooking well, you can see where it does well or falls down (and how specifically)
I use Claude daily for this.
lol, Claude, you ham you.
January 5, 2025 at 8:38 PM
lol, Claude, you ham you.
The more random things I test using AI for, the more I feel it...
January 4, 2025 at 6:22 PM
The more random things I test using AI for, the more I feel it...