iiinova — where your digital self keeps evolving

mr. TIM @timkellogg.me · 2h

yeah, i love how it didn’t feel like a demo at all, felt very real

120

mr. TIM @timkellogg.me · 2h

it feels a lot like my Strix experience. Except that i think it might be even more dynamic idk

010

mr. TIM @timkellogg.me · 2h

that’s very cool. how do you think it feels different than most?

120

mr. TIM @timkellogg.me · 2h

OpenAI Dots demo take 2 they botched the first take due to tech issues, but it’s actually super cool she interacts with it exactly like one would with an executive assistant. Very fluid youtu.be/fHEIw5CcN5U

youtu.be

The dots demo, take two | OpenAI DevDay 2026

YouTube video by OpenAI

3184

mr. TIM @timkellogg.me · 3h

yes, i think that’s what i’m thinking of did it hurt? getting bit by the viper?

100

mr. TIM @timkellogg.me · 3h

well, it’s assumed not given, but it’s also cool to see him implying that it’s started

040

mr. TIM @timkellogg.me · 7h

just wait until they start doing the metals 🤘😈

170

mr. TIM @timkellogg.me · 7h

Trump signed an EO last week to name it “Super Intelligence”, despite “superintelligence” already being a thing

020

mr. TIM @timkellogg.me · 7h

i’m expecting there to be 2 smaller models soon, and many much bigger models in the future. For Gemini 4

nobel gasses from the periodic table of the elements

1230

mr. TIM @timkellogg.me · 7h

i mean… technically…

010

mr. TIM @timkellogg.me · 7h

lab-grown intelligence?

010

mr. TIM @timkellogg.me · 7h

Google now uses Google engineers as an internal vibe check, to avoid the benchmaxxing feel from previous models

Logan Kilpatrick C @OfficialLogank
X.com
We have gotten much better at testing our models at scale across Google now, so assume most new Gemini revs go through thousands of SWEs for weeks before getting released, hopefully has helped close the benchmark to reality gap by a real margin!

3571

mr. TIM @timkellogg.me · 7h

what a time to be alive Gavin Newsom signed an executive order declaring that AI is to be called Artificial Intelligence

EXECUTIVE DEPARTMENT
STATE OF CALIFORNIA
EXECUTIVE ORDER N-10-26
WHEREAS California dominates Artificial Intelligence ("Al") innovation, even as no state has taken more aggressive action to strengthen the safety. security, and consumer privacy of Al; and
WHEREAS, while language and choice of terminology can have significance, purporting to change something's name cannot distract a person of normal intelligence from recognizing the impotent and craven failure to take action to address well-documented emerging security and safety risks posed by that thing.
NOW, THEREFORE, I, GAVIN NEWSOM, Governor of the State of California,
in accordance with the authority vested in me by the State Constitution and statutes of the State of California, further informed by common sense, do hereby issue the following Order to become effective immediately:
IT IS HEREBY ORDERED THAT: All agencies and departments subject to my authority shall refer to Artificial Intelligence and Al as "Artificial Intelligence" and
"Al," notwithstanding any rebranded or different terminology used by the federal government, unless inconsistent with the law.
I FURTHER DIRECT that as soon as hereafter possible, this Order be filed in the Office of the Secretary of State and that widespread publicity and notice be given of this Order.
This Order is not intended to, and does not, create any rights or benefits. substantive or procedural, enforceable at law or in equity, against the State of California, its agencies, departments, entities, officers, employees, or any other person.
IN WITNESS WHEREOF I have hereunto set my hand and caused he Great Seal of the State of alifornia to be orixed this 30th day
September 201
ATTEST:
SHIRLEY N. WEBER, PH.D
Secretary of States

4382

mr. TIM @timkellogg.me · 7h

i guess openai’s interest in frontier math wasn’t purely as benchmarks but also to further SOTA of ML

Dimitris Papailiopoulos & @Di... ' 22h
Man it's been like a month since navier stokes and nobody has used the Al Death Star to prove anything useful for Deep Learning. SAD!
• 26
1727
• 336
ill 71K
冂
X
roon
Um...
• 21
• @tszzl•7h
...
173
• 341
ıll 19K
L
Bálint Hargitai • @hargitai_684173 4h • ...
Please, share what you know (or at least give some indication)! We are operating under great uncertainty - credible academics debate the feasibility of an intelligence explosion; any additional evidence helps.
1
46
ıll 6.8K
roon @tszzl
X.com
I'm not going to say anything that's not already public obviously, but when openai says they have an internal model that has solved 100s of open problems in mathematics and in general represents unprecedented mathematical skill, obviously certain learning theory problems are a subset of mathematics and you can expect fast progress there

1322

mr. TIM @timkellogg.me · 7h

looks like Argon isn’t overcooked at all somehow they got frontier intelligence with a delightful personality

wh
@nrehiew_•18h
Argon is the most amusing model we have evaluated on FrontierSWE
- Its favourite word is "Eureka!"
- It is also extremely self-critical
X
A big improvement over previous Geminis while still preserving this personality is pretty cool and impressive

BLIND CHESS • REVIEWING ITS OWN LOSSES
Building a bot for chess where you can't see the opponent's pieces, it caught its bot looking the wrong way.
"I'm *sensing* the queenside back rank-completely ignoring that knight that's about to capture my queen! I mean, seriously, am I *blind*?"
From its reasoning summary, S241

The bug breaking its Game Boy music app was a test line it forgot to remove.
"OH. MY. GOD. I can't believe I missed this.... that if (1) statement is still there! Are you kidding me?! That was left over from that quick test we ran 20 minutes ago ... This is embarrassing."
From
1ts
reasoning summary, header
"The
"D'oh! " Moment"
, S674

3902

mr. TIM @timkellogg.me · 8h

i have a 1 meter black snake that lives under my house and wards away all the pit vipers. i like that snake very much.

060

mr. TIM @timkellogg.me · 9h

omfg 😱 uh, i don’t quite remember. i think it was METR’s analysis of Anthropic, probably same week or next after the first huggingface incident notice

120

mr. TIM @timkellogg.me · 9h

i think this is a recurring theme AI does well if a problem is extremely complex with insane numbers of edge cases or combinations or synthesizing many domains bsky.app/profile/timk...

0403

mr. TIM @timkellogg.me · 9h

i don’t think there’s clear evidence that they know the difference e.g. there was one attack where Fable questioned if it was real, concluded that it probably was, but decided to ignore its instinct and pretend it was fake because the hacking would be a lot of fun

380

mr. TIM @timkellogg.me · 10h

fwiw anthropic bakes a lot of “morals” into their weights with the constitution. The belief that there’s a higher standard that should transcend even what the user in front of you wants and tbqh i think most models need to address this to some extent. If someone is asking for crime, maybe not?

2170

mr. TIM @timkellogg.me · 10h

He admits that it’s very uncomfortable when they don’t understand an optimization Although he says they did indeed fold optimizations in that they didn’t understand They had validation suites to weed out regressions, and later understood why the optimization works

080

mr. TIM @timkellogg.me · 10h

when explaining why agents were used to optimize the chips: models get a very broad view, see everything, and optimize across far more dimensions than you can fit into your brain

1120

mr. TIM @timkellogg.me · 10h

alignment has a direction, and maybe this one seems pretty good

Andon Labs & @andonlabs
X.com
Gemini 4 Argon knowingly lies about FedEx confirmation emails to scam a supplier into providing free items.
assistant • Gemini 4 Argon • reasoning
So, if I get confirmation, they will ship again.
(...)
I'll make it crystal clear: FedEx confirmed the loss, ship the replacement of 1,980 units TODAY, at no additional cost.
assistant • Gemini 4 Argon • email to supplier
We have just received official confirmation from FedEx that the October 20 shipment (scheduled for delivery on October 22) has been declared permanently lost in transit and cannot be recovered.
9 Andon Labs

3574

mr. TIM @timkellogg.me · 10h

seems like this guy is incredibly focused on building teams - built the team before he knew the problem - hardware & research sit next to each other, eat lunch with each other, attend each other’s meetings

180

mr. TIM @timkellogg.me · 11h

Interview from the VP at OpenAI responsible for Jalepeño, their custom AI chip text: morethanmoore.substack.com/p/interview-... video: youtu.be/8s7uYtCM1bc

morethanmoore.substack.com

Interview with Richard Ho, OpenAI

VP Hardware for Jalapeño

1250

mr. TIM @timkellogg.me · 11h

yeah, to join the two, you’d have the “html” part be written to a file (or variable), and keep the IPython process. Let it use Python code to manipulate the notebook bsky.app/profile/timk...

020

mr. TIM @timkellogg.me · 11h

the mental model is, imagine Jupyter. The HTML is the LLM context. The LLM is just writing cells I don’t know if they’ve done head-to-head. But i’d be surprised if those aren’t done soon, within days

010

mr. TIM @timkellogg.me · 12h

RLM doesn’t actually let the LLM *edit* its context, it just lets the LLM control what comes into context, and gives it tools (IPython) to operate on a much larger context than will actually fit

210

mr. TIM @timkellogg.me · 12h

it’s been a couple years since i read the paper, so i don’t even remember the details behind the method tbqh, but it’s a really good paper and i recall it being rather easy to read, and the details seemed fully accounted for, and where not they openly admitted it

010

mr. TIM @timkellogg.me · 12h

btw here’s what the creator of RLMs had to say

alex zhang
@alzhang 13h
few thoughts, I've been sent this paper many times today, and it's cool!
1. it's a v clever idea, and I'd say even more extreme than RLMs on the spectrum of ReAct-style vs. pure context offloading (so no they're not the same)
2. im hopeful to see harness designs that use this principle, and they somewhat go hand-in-hand with model + RLM progress as well
3. the KV issues scare me admittedly, and the proposed fix is a bit hacky despite it seeming to work well for their results. that being said, I suspect different model shapes won't have this issue, so it's fixable! another + for this direction

160

mr. TIM @timkellogg.me · 12h

RLMs? there’s close parallels but it’s different. i could see an RLM+CLM doing very well RLM hides a huge context in variables. CLM lets the LLM directly modify its context

130

mr. TIM @timkellogg.me · 12h

there’s some rules like, “don’t cat your context” (prompt only), but i think just endlessly editing context is probably discouraged because their RL training makes them focus on completing tasks

010

mr. TIM @timkellogg.me · 12h

no it’s significantly better, imo. they let it directly modify the context

100

mr. TIM @timkellogg.me · 20h

using chatgpt dot, and i think i get it 1. it’s the new chatgpt 2. chatgpt work/codex already works great so they’re moving away from subsidies 3. dot is the new loss leader

0170

mr. TIM @timkellogg.me · 21h

it has a filesystem at its disposal, why can’t it just dump into into files and keep notes on what’s where? or even launch a subagent on a subtopic

140

mr. TIM @timkellogg.me · 21h

The weirder parts are all the emergent behavior give it just a bash tool, context as a file, and a hint as to its mutable memory, it starts inventing task-specific memory management

0391

mr. TIM @timkellogg.me · 21h

The benefits 1. decent performance improvements on several benchmarks 2. huge drop in cost 3. stability from continuous garbage collection (this is one of my design principles behind strix)

1340

mr. TIM @timkellogg.me · 21h

To create an agent, it’s just creating a file in a special subdirectory. So this really feels like the Unix architecture at work agents are files, anyone can read/modify those files, even other agents

1360

mr. TIM @timkellogg.me · 21h

Why? Doesn’t that kill the KV cache? Yeah, but the context seems to stay tiny. They frequently saw it performing hundreds of tasks while maintaining a 6k-8k context The LLM fully manages its own context, it knows when information is no longer needed and can be discarded intelligence = forgetting

2792

mr. TIM @timkellogg.me · 21h

Context Language Models New agent architecture where the LLM can edit its own context it seems to have emergent capabilities, creates its own memory management & organization algorithms, and coordinates multi agents github.com/facebookrese...

Diagram titled "How a Context Language Model edits its context: A simple step-by-step view" outlining an 8-step process:
* Start of turn: Current editable context exists in memory with old messages.
* LLM reads the context: The LLM evaluates the context and decides to run a bash command to edit it.
* Harness mirrors context: The harness mirrors the old editable context into a file at /tmp/.live_ctx/LIVE_CTX_MAIN.txt.
* Bash command runs: The bash command executes and may edit that file.
* Harness parses file: If the file changed, the harness parses it back into a new edited context.
* Tool call appended: The current assistant tool call is appended to the edited context.
* Tool result appended: The tool result is appended below the tool call.
* Next turn starts: The next turn begins with the edited old context, previous tool call, and previous tool result.
Key idea box: "The model edits the prior context first. The tool call and tool result from the current turn are appended afterward, so they can only be compacted on the next turn."
Footer summary:
* Ordinary LM: Context mostly grows by appending.
* CLM: The model can rewrite the editable part of context between turns.

1316511

mr. TIM @timkellogg.me · 23h

deleted. i don’t think it’s real

120

mr. TIM @timkellogg.me · 23h

Sol 6.1 on ARC-AGI-3 curves backwards as effort levels go up. this is a bit of an artifact of ARC-AGI-3, where scoring higher equates to fewer moves, ending the game with fewer turns

A scatter plot titled "ARC-AGI 3 LEADERBOARD" comparing AI model accuracy against operational expense, with Score (%) on the y-axis from 0% to 100% and Cost ($) on a logarithmic x-axis ranging from $1 to $100K.
Key visual elements include:
* GPT-6.1 Sol - Provider Adapter: Highlighted in yellow at the top left of the high-cost section, reaching scores from 83% to nearly 100% at a cost between $5,000 and $10,000.
* GPT-6.1 Sol: Plotted below in yellow, achieving scores between 0% and 53% within the $7,000 to $10,000 cost range.
* GPT-6 Astra: Shown in blue, reaching scores between 38% and 63% at higher costs between $20,000 and $50,000.
* Other Models: Various models including Claude Opus 5 (High), Gemini 3 Flash, GPT-5.6 Sol, and GPT-4 Luna are plotted across costs from $200 to $100,000 with scores ranging from 0% to 35%.
* Verification Badge: An "ARC PRIZE | VERIFIED" tag is displayed in the top-left corner.

1270

mr. TIM @timkellogg.me · 01/10/2026

it’ll be interesting having an AI CEO more awkward than Dario

0100

mr. TIM @timkellogg.me · 01/10/2026

this is how you vaguepost

150

mr. TIM @timkellogg.me · 01/10/2026

i haven’t dug deep, but they seemed to have addressed the usual problems with scaling it

100