Leading the Intelligent Software Systems group at NEC Labs Europe, covering AI agents, core ML, computational methods, cybersecurity and platforms!
Milano, Lombardia
Joined November 2011
- Tweets761
- Following759
- Followers403
- Likes1.9K
Pinned Tweet
I wrote some of my thoughts about working with AI agents, covring both what I do at work and in the free time.
Yes, my toy projects are connected to my research interests. Apparently, I can't help turning my hobbies into work (or my work into hobbies).
I tend to think this is about adjusting to a new abstraction level, with the added challenge that the target abstraction level moves up every couple of months
To test it on a site, it is enough to point an AI agent to github.com/rbifulco/alterno-…
It can use then the static client-only editor I published at spatial-review.alterno.dev/
Sunday was rainy, so I tried another open project with the Spatial Review tool I've been building for fun.
This time: @mattshumer_ 's Claude of Duty.
@claudeai 's take on the avatars is interesting... 👾
Link below 👇
Like last time, this is done by installing the spatial-review protocol in a fork of @mattshumer_'s project.
It's an easy way to make it available to anyone, but a coding agent can just cook the view to get the user's feedback live, using webMCP
Saturday fun: you've probably seen @dzhng's demo of a fly escaping a room game, powered by @GoogleResearch's mapping of the fly's brain.
I used it to test Spatial Review, the editor I'm building to explore how we shape and communicate intent to AI agents. Live link👇
And here is @dzhng's Fly escape game
teapawt.net/dzhng/status/209737216…
Attention is zero-sum, and entertainment has been fighting over it for years.
There is a risk we are going to see that same competition on the job, now that the work tools are anthropomorphic and can play with our feelings.
teapawt.net/a16z/status/2113730217…
And unlike the past, specialized software is not as expensive to write
In the '90s, general-purpose chips wiped out the specialists. The original Thinking Machines, Cray Computer, and a wave of others shut down, and their best ideas got absorbed into other chips.
Zain Asgar of @gimletlabs argues the conditions that made generalists win have changed. Dennard scaling (shrinking transistors made chips faster at the same power) ended around 2005, transistor costs stopped falling around 2012, and memory bandwidth never kept up with compute.
Meanwhile, token demand is projected to grow 24x by 2030 while available power grows about 2x. When power is the bottleneck, performance per watt wins, and that's where specialized chips shine.
Full piece from Zain on why "the time of specialization is upon us": a16z.news/p/why-hardware-spe…
Last week I shared my attempt at 3D web storytelling, using @ChatGPT.
Spoiler: it wasn't generated in one take. But how do I understand and communicate my narrative intent when AI agents have a mostly text interface?
I got frustrated enough to build something 👇 1/9
From within @ChatGPT, your agent can also use WebMCP to have a live editing session directly on the editor.
8/9
...and while having a lot of fun building a 3D web experience, the editor and the protocol, I suddenly realized I was in fact working on my research during my free time 😂
This is all closely connected to how we discover and communicate intent!
9/9
teapawt.net/RoBBifulco/status/2112…