Provost AI Fellow. ECE Prof @uiowa. AI + VR/AR + 3D + HCI researcher at the intersection of AI, education, telepresence, accessibility, & creativity. Optimist.
Iowa City, IA
Joined October 2009
- Tweets8.4K
- Following1.4K
- Followers1.1K
- Likes7.4K
Can't wait to cover this in my IoT course in the spring.
“Our experiments suggest that small AI models may be months away from being capable enough to self-replicate on commercial IoT devices, like smart ovens and off-brand Android TV boxes. A replicator like this would be challenging to shut down, since there’s no centralized point to shut off - though it may be slow (with a doubling time of days). Unlike non-AI malware, it could also figure out unfamiliar devices on its own. In our evals, a 9B model got inference running on 8 of the 9 hardware types we tested.
Models of 30B parameters and under can already complete many of the individual steps for replication today, though they make mistakes when composing them end to end. In our simplified evals, GLM-5.2 and stronger models are capable of end to end replication. If small models improve at their historical rate, they’ll match this capability level by the end of this year. Small weight-based replicators may be possible soon after.” -@bucketofkets @willzhou97
anthropic.com/research/self-…
Tyler Bell retweeted
I put Sonic Adventure into GTAV and I literally can't stop playing it
Sonic, Tails and Knuckles are all fully playable
This is the golden age of modding
I made this slide in 2023. The proactivity of agents like dots and Muse are showing signs of this ubiquitous, always-on nature.
Example: my dot pinged me before EOD yesterday asking if I’d yet done a task I’d mentioned on Tuesday to see if I needed help.
An exciting shift!
Delegate work in Codex to your dot.
Your dot can now start work in Codex and follow up on existing threads, drawing on your ChatGPT conversations, Codex threads, and automations. It’s also better at deciding when to keep a thread going or start fresh.
Your dot can also review and edit Scheduled Tasks in ChatGPT Work, too. Ask what’s scheduled or update a recurring task as plans change.
Tomorrow’s lecture kicks off a mini-series on AI control, alignment, and human-AI interaction. Tonight I have Claude cooking up some (very compelling) supplementary materials, this one about how thought experiments like paperclip maximizer are becoming increasingly pertinent 🖇️
From “Understand” by Ted Chiang (1991)
What happens when a single day of AI research produces more mathematics than humanity can absorb in a century?
Imagine millions of superhuman research agents at work. Formal systems may verify the proofs, but humans won’t have enough context to understand the underlying web of machine-invented concepts.
Mathematics could split into two layers:
Machine mathematics: vast, verified, mostly consumed by AIs.
Human mathematics: a compressed “effective theory” of the machine frontier—the small subset of ideas we can understand.
Physics may follow the same path: AIs could explore huge spaces of theories and derive correct predictions using structures no human physicist fully understands.
The key question is whether AI can compress knowledge as fast as it creates it.
My guess: the detailed frontier becomes permanently machine-native, while selected parts are periodically compressed into human-comprehensible principles.
Eventually even deciding what is worth understanding may require AI.
At that point AI is no longer just a scientific tool. Humans are receiving selected explanations from a much larger intellectual civilization they cannot independently grasp.
Ted Chiang essentially predicting in 1991 what working with agents in Codex and Claude Code would be like 😂
(For real though “Understand” by Chiang is so compelling)
Having my dot use Blender on its own cloud machine to design faceplates to 3D print for my @pollenrobotics Reachy Mini
I finally get to show this image in lecture today as we broaden our LLM discussions toward Agents and harnesses.
(if anyone knows the OP author / post, please lmk)
Reachy performs the YMCA! 🕺
Opus 5.5 implemented all of the choreography, I just had to play dance instructor for a few parts!
@pollenrobotics
I’m letting Opus 5.5 create a dance routine to “YMCA” for my Reachy robot. Watching these models check their work never ceases to amaze me. Here Claude is balancing the dance “vibe” with the timing of the lyrics (which it extracted and timestamped itself) with the safety / feasibility of actually reaching these target positions without breaking the hardware. Incredible.
I’m letting Opus 5.5 create a dance routine to “YMCA” for my Reachy robot. Watching these models check their work never ceases to amaze me. Here Claude is balancing the dance “vibe” with the timing of the lyrics (which it extracted and timestamped itself) with the safety / feasibility of actually reaching these target positions without breaking the hardware. Incredible.
I asked the new Claude Sonnet 5.5 to draw a self-portrait using only JavaScript in a variety of styles.
Here’s typographic. It came with the caption, “Made up of words I spend my days on: asking, weighing, wondering, answering.”
