Tim Kellogg
@timkellogg.me
AI Architect | North Carolina | AI/ML, IoT, science
WARNING: I talk about kids sometimes
WARNING: I talk about kids sometimes
Pinned
Social Media “Nutrition Label” for me for the last several days (thanks nano banana!)
towards the end, Ilya has a part where he makes the case that the genome (thus evolution) doesn't dictate intelligence
i have a hunch that that's why he's taking a lot of crap from some parts of the tech bro crowd that's started leaning into eugenics. Might have nothing to do with his AI views
i have a hunch that that's why he's taking a lot of crap from some parts of the tech bro crowd that's started leaning into eugenics. Might have nothing to do with his AI views
November 26, 2025 at 12:56 PM
towards the end, Ilya has a part where he makes the case that the genome (thus evolution) doesn't dictate intelligence
i have a hunch that that's why he's taking a lot of crap from some parts of the tech bro crowd that's started leaning into eugenics. Might have nothing to do with his AI views
i have a hunch that that's why he's taking a lot of crap from some parts of the tech bro crowd that's started leaning into eugenics. Might have nothing to do with his AI views
this is insane. how small can we go?
the even cooler part is this all independent research
the even cooler part is this all independent research
The threshold for consistent English/query understanding is now 3M parameters.
November 26, 2025 at 11:58 AM
this is insane. how small can we go?
the even cooler part is this all independent research
the even cooler part is this all independent research
new summary of the Ilya podcast just landed
If he was in a band, he would be saying that this next album is going to suck but it’s “getting back to our roots” kind of suck.
November 26, 2025 at 11:53 AM
new summary of the Ilya podcast just landed
thinking more about Ilya’s strategy..
that’s why he’s on a podcast, to shape minds. he can’t just release a shitty model and be called a saint. he needs to control the narrative and provide context for what he’s done
if this doesn’t land, he’s likely screwed (ngl i don’t think it landed)
that’s why he’s on a podcast, to shape minds. he can’t just release a shitty model and be called a saint. he needs to control the narrative and provide context for what he’s done
if this doesn’t land, he’s likely screwed (ngl i don’t think it landed)
yeah, his idea is really like releasing a newborn baby and saying, “that’s it, work is done”
but it’s not done, it’s still got to learn
in our current approaches, it’s hard to conceive of that, because we’re bombarded by hype and marketing. i can’t imagine releasing an incapable model..
but it’s not done, it’s still got to learn
in our current approaches, it’s hard to conceive of that, because we’re bombarded by hype and marketing. i can’t imagine releasing an incapable model..
November 26, 2025 at 2:52 AM
thinking more about Ilya’s strategy..
that’s why he’s on a podcast, to shape minds. he can’t just release a shitty model and be called a saint. he needs to control the narrative and provide context for what he’s done
if this doesn’t land, he’s likely screwed (ngl i don’t think it landed)
that’s why he’s on a podcast, to shape minds. he can’t just release a shitty model and be called a saint. he needs to control the narrative and provide context for what he’s done
if this doesn’t land, he’s likely screwed (ngl i don’t think it landed)
Summary — He's got a divergent view of AGI
We're all pursuing a single behemoth that is *already* smarter than all humans when it's launched
He's pursuing an entity that is *capable of* being smarter
i.e. he's all in on continual learning
We're all pursuing a single behemoth that is *already* smarter than all humans when it's launched
He's pursuing an entity that is *capable of* being smarter
i.e. he's all in on continual learning
November 26, 2025 at 2:02 AM
Summary — He's got a divergent view of AGI
We're all pursuing a single behemoth that is *already* smarter than all humans when it's launched
He's pursuing an entity that is *capable of* being smarter
i.e. he's all in on continual learning
We're all pursuing a single behemoth that is *already* smarter than all humans when it's launched
He's pursuing an entity that is *capable of* being smarter
i.e. he's all in on continual learning
alright, Opus 4.5 got me through a gnarly debugging session (not one shot) that Gemini 3 couldn't figure out
Opus 4.5 basically does not do doom loops, period. It's legit, I'm impressed.
Opus 4.5 basically does not do doom loops, period. It's legit, I'm impressed.
November 25, 2025 at 10:41 PM
alright, Opus 4.5 got me through a gnarly debugging session (not one shot) that Gemini 3 couldn't figure out
Opus 4.5 basically does not do doom loops, period. It's legit, I'm impressed.
Opus 4.5 basically does not do doom loops, period. It's legit, I'm impressed.
oh shit, NVIDIA’s in trouble
November 25, 2025 at 9:45 PM
oh shit, NVIDIA’s in trouble
Ilya!!!!
www.dwarkesh.com/p/ilya-sutsk...
www.dwarkesh.com/p/ilya-sutsk...
Ilya Sutskever – We're moving from the age of scaling to the age of research
“These models somehow just generalize dramatically worse than people. It's a very fundamental thing.”
www.dwarkesh.com
November 25, 2025 at 6:07 PM
Ilya!!!!
www.dwarkesh.com/p/ilya-sutsk...
www.dwarkesh.com/p/ilya-sutsk...
too many people seem to be convinced that LLM vendors set prices on a cost plus basis
no, the advantage of closed weights is you can explore prices completely detached from cost. You’re free to set prices based purely on what people will pay, the value they get from it
no, the advantage of closed weights is you can explore prices completely detached from cost. You’re free to set prices based purely on what people will pay, the value they get from it
November 25, 2025 at 3:29 PM
too many people seem to be convinced that LLM vendors set prices on a cost plus basis
no, the advantage of closed weights is you can explore prices completely detached from cost. You’re free to set prices based purely on what people will pay, the value they get from it
no, the advantage of closed weights is you can explore prices completely detached from cost. You’re free to set prices based purely on what people will pay, the value they get from it
codex just taught me about jina.ai reader
an API you can easily use via curl that takes a URL and converts it to LLM-friendly text. Free to use, afaict
github.com/jina-ai/reader
an API you can easily use via curl that takes a URL and converts it to LLM-friendly text. Free to use, afaict
github.com/jina-ai/reader
GitHub - jina-ai/reader: Convert any URL to an LLM-friendly input with a simple prefix https://r.jina.ai/
Convert any URL to an LLM-friendly input with a simple prefix https://r.jina.ai/ - jina-ai/reader
github.com
November 25, 2025 at 2:20 PM
codex just taught me about jina.ai reader
an API you can easily use via curl that takes a URL and converts it to LLM-friendly text. Free to use, afaict
github.com/jina-ai/reader
an API you can easily use via curl that takes a URL and converts it to LLM-friendly text. Free to use, afaict
github.com/jina-ai/reader
calling out bs on my own posts because microsoft can’t be trusted to not commit chart crimes
community note: using cost on the y axis makes it appear like cheaper models are more capable on pass@3
November 25, 2025 at 2:10 PM
calling out bs on my own posts because microsoft can’t be trusted to not commit chart crimes
one side conversation i had multiple times at AIE was that maybe mono repos are good
if all of your dependencies are sitting on disk, the agent doesn’t need to rely on documentation
even wo monorepos, it’s a good idea to clone tricky dependencies locally
if all of your dependencies are sitting on disk, the agent doesn’t need to rely on documentation
even wo monorepos, it’s a good idea to clone tricky dependencies locally
November 25, 2025 at 2:04 PM
one side conversation i had multiple times at AIE was that maybe mono repos are good
if all of your dependencies are sitting on disk, the agent doesn’t need to rely on documentation
even wo monorepos, it’s a good idea to clone tricky dependencies locally
if all of your dependencies are sitting on disk, the agent doesn’t need to rely on documentation
even wo monorepos, it’s a good idea to clone tricky dependencies locally
Fara 7B: A cheap & capable open weights computer use agent (CuA)
they got within a few points of o3’s performance using only 4k training data points (yes, synthetic)
www.microsoft.com/en-us/resear...
they got within a few points of o3’s performance using only 4k training data points (yes, synthetic)
www.microsoft.com/en-us/resear...
November 25, 2025 at 1:54 PM
Fara 7B: A cheap & capable open weights computer use agent (CuA)
they got within a few points of o3’s performance using only 4k training data points (yes, synthetic)
www.microsoft.com/en-us/resear...
they got within a few points of o3’s performance using only 4k training data points (yes, synthetic)
www.microsoft.com/en-us/resear...
Exa 2.1: both fast and accurate search (that’s not Google)
available both as an MCP server & web UI
exa.ai/blog/exa-api...
available both as an MCP server & web UI
exa.ai/blog/exa-api...
November 25, 2025 at 1:42 PM
Exa 2.1: both fast and accurate search (that’s not Google)
available both as an MCP server & web UI
exa.ai/blog/exa-api...
available both as an MCP server & web UI
exa.ai/blog/exa-api...
this is the value of these new scaled models
GPT-5-Pro could probably do it too, but you’d pay like $30 for one shot
Gemini 3 & Opus 4.5 can still run fast & cheap bc they’re extremely sparse MoE, but solve very tricky problems
we truly need scale along both axes
GPT-5-Pro could probably do it too, but you’d pay like $30 for one shot
Gemini 3 & Opus 4.5 can still run fast & cheap bc they’re extremely sparse MoE, but solve very tricky problems
we truly need scale along both axes
Opus 4.5 solved a very tricky, complex problem in one session for me (VS Code Agent mode) that Sonnet 4.5 had been giving up on all day yesterday (I'm quite relentless).
November 25, 2025 at 12:48 PM
this is the value of these new scaled models
GPT-5-Pro could probably do it too, but you’d pay like $30 for one shot
Gemini 3 & Opus 4.5 can still run fast & cheap bc they’re extremely sparse MoE, but solve very tricky problems
we truly need scale along both axes
GPT-5-Pro could probably do it too, but you’d pay like $30 for one shot
Gemini 3 & Opus 4.5 can still run fast & cheap bc they’re extremely sparse MoE, but solve very tricky problems
we truly need scale along both axes
that’s it, i’m calling it, software engineering is over
AI can do everything an engineer can do
AI can do everything an engineer can do
![80×24
user@machine:~$ codex-cli execute --parallel=10 -verbose --mode=chaos
[09:15:221
[CLI] Initializing 10 parallel workers on local GPU cluster...
[09:15:23]
[CLI] Loaded model: codex-davinci-003-turbo
[09:15:23]
[CLI] WARNING: 'chaos' mode active. Expect non-deterministic output.
LPID
3101]
[TASK 01]
[PID
• START: Refactoring legacy Python 2.7 codebase to idiomatic Rust.
3102]
[TASK 02]
RUNNING: Generating 500 unit tests for an empty function. [==>..••...] 25%
IPID
3103]
[TASK 03]
START: Explaining Kubernetes architecture using only food analogies.
IPID
3104]
[TASK 04]
DONE: Optimized a Dockerfile to be 12 bytes smaller. (took 0.2s)
[PID 3105]
[TASK 05]
.........•]
5%
RUNNING: Translating the entire Linux kernel comments into Haiku form. >...
[PID 3106] [TASK 06] • START: Attempting to vertically center a div using CSS and machine learning.
[PID 3107] [TASK 07] loop detected).
A STALLED: Arguing with another GPT instance about tabs vs. spaces. (Infinite
[PID 3108]
LTASK 08]
>........•] 1%
[PID 3109] ITASK 09]
RUNNING: Trying to exit Vim using reinforcement learning. Iteration 2,491.
CRITICAL: Attempting to parse HTML with regex. Elder Gods summoned. ZALGO
imminent.
[PID 3110]ITASK 10] • START: Predicting the exact heat death of the universe based on current npm dependency trees.
[09:15:35] [CLI] STATUS: 1 Finished, 4 Running, 1 Stalled, 1 CRITICAL, 3 Pending. GPU Temp: 92°C.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:usglohqysqawyhpspgg3hemf/bafkreia2gvuyfom2mwcl3tufcpedzom6x2bjhnbzwqrmjtu3nuftg5yxtq@jpeg)
November 25, 2025 at 12:23 PM
that’s it, i’m calling it, software engineering is over
AI can do everything an engineer can do
AI can do everything an engineer can do
Reposted by Tim Kellogg
Opus 4.5 solved a very tricky, complex problem in one session for me (VS Code Agent mode) that Sonnet 4.5 had been giving up on all day yesterday (I'm quite relentless).
November 25, 2025 at 7:32 AM
Opus 4.5 solved a very tricky, complex problem in one session for me (VS Code Agent mode) that Sonnet 4.5 had been giving up on all day yesterday (I'm quite relentless).
merging the kiddo’s trio of passions for space facts, hamsters, and K-pop demon hunter
November 25, 2025 at 12:44 AM
merging the kiddo’s trio of passions for space facts, hamsters, and K-pop demon hunter
i built this for myself a few months ago. it worked well, except that i only launched them in subagents (to preserve the prefix cache). this would probably work a lot better
no such thing as too many tools!
no such thing as too many tools!
A tool for searching for relevant tools to keep context clean?
Was thinking about this last night as I approached sleep and glad to find this morning that one of the thought leaders rolled out this capability
www.anthropic.com/engineering/...
Was thinking about this last night as I approached sleep and glad to find this morning that one of the thought leaders rolled out this capability
www.anthropic.com/engineering/...
Introducing advanced tool use on the Claude Developer Platform
Claude can now discover, learn, and execute tools dynamically to enable agents that take action in the real world. Here’s how.
www.anthropic.com
November 25, 2025 at 12:39 AM
i built this for myself a few months ago. it worked well, except that i only launched them in subagents (to preserve the prefix cache). this would probably work a lot better
no such thing as too many tools!
no such thing as too many tools!
oooh, codex-cli is doing 2 things at once
did it start supporting subagents? i missed that
did it start supporting subagents? i missed that
November 24, 2025 at 10:44 PM
oooh, codex-cli is doing 2 things at once
did it start supporting subagents? i missed that
did it start supporting subagents? i missed that
i reeeally hope this is what it looks like
i’d love to hear from Ilya, and also i assume Ilya wouldn’t talk unless he had something interesting to say, some tidbit of news also dropping tomorrow
i’d love to hear from Ilya, and also i assume Ilya wouldn’t talk unless he had something interesting to say, some tidbit of news also dropping tomorrow
November 24, 2025 at 10:18 PM
i reeeally hope this is what it looks like
i’d love to hear from Ilya, and also i assume Ilya wouldn’t talk unless he had something interesting to say, some tidbit of news also dropping tomorrow
i’d love to hear from Ilya, and also i assume Ilya wouldn’t talk unless he had something interesting to say, some tidbit of news also dropping tomorrow
OpenAI is planning on releasing 2 models in the next few months:
- GPT-5.2: Successor, very good at programming
- Shallotpeat: fixed pre-training + new base for the IMO Gold math model
I'm really curious about Shallotpeat. Sounds like a redo of GPT-4.5
- GPT-5.2: Successor, very good at programming
- Shallotpeat: fixed pre-training + new base for the IMO Gold math model
I'm really curious about Shallotpeat. Sounds like a redo of GPT-4.5
November 24, 2025 at 10:10 PM
OpenAI is planning on releasing 2 models in the next few months:
- GPT-5.2: Successor, very good at programming
- Shallotpeat: fixed pre-training + new base for the IMO Gold math model
I'm really curious about Shallotpeat. Sounds like a redo of GPT-4.5
- GPT-5.2: Successor, very good at programming
- Shallotpeat: fixed pre-training + new base for the IMO Gold math model
I'm really curious about Shallotpeat. Sounds like a redo of GPT-4.5
its both exciting and slightly frustrating that Opus 4.5 is both better and worse than Gemini 3 Pro
Opus => Coding
Gemini => Problem solving, explaining
Opus => Coding
Gemini => Problem solving, explaining
November 24, 2025 at 10:06 PM
its both exciting and slightly frustrating that Opus 4.5 is both better and worse than Gemini 3 Pro
Opus => Coding
Gemini => Problem solving, explaining
Opus => Coding
Gemini => Problem solving, explaining
Opus 4.5 scored 50% higher than Gemini 3 Pro on the “system card page count” benchmark
system card
assets.anthropic.com/m/64823ba748...
oh, high alignment and low rates of concerning behavior? sounds like bliss
assets.anthropic.com/m/64823ba748...
oh, high alignment and low rates of concerning behavior? sounds like bliss
November 24, 2025 at 8:30 PM
Opus 4.5 scored 50% higher than Gemini 3 Pro on the “system card page count” benchmark