@braintelligence.bsky.social
Believer in inclusive democracy
Posting mostly about AI/ML and tech if I can help it
Posting mostly about AI/ML and tech if I can help it
Reposted
The number of companies that can land a rocket appears to have doubled
November 13, 2025 at 10:46 PM
The number of companies that can land a rocket appears to have doubled
Reposted
Weibo, China's Twitter, released VibeThinker-1.5B — SOTA reasoning in a tiny model.
🚀 Performance: Highly competitive on AIME24/25 & HMMT25 — surpasses DeepSeek R1-0120 on math, and outperforms same-size models in competitive coding.
🚀 Performance: Highly competitive on AIME24/25 & HMMT25 — surpasses DeepSeek R1-0120 on math, and outperforms same-size models in competitive coding.
November 13, 2025 at 6:23 AM
Weibo, China's Twitter, released VibeThinker-1.5B — SOTA reasoning in a tiny model.
🚀 Performance: Highly competitive on AIME24/25 & HMMT25 — surpasses DeepSeek R1-0120 on math, and outperforms same-size models in competitive coding.
🚀 Performance: Highly competitive on AIME24/25 & HMMT25 — surpasses DeepSeek R1-0120 on math, and outperforms same-size models in competitive coding.
Reposted
"Solving a Million-Step LLM Task with Zero Errors", from the inimitable Elliot Meyerson (et al.)
Solving a Million-Step LLM Task with Zero Errors
LLMs have achieved remarkable breakthroughs in reasoning, insights, and tool use, but chaining these abilities into extended processes at the scale of those routinely executed by humans, organizations...
arxiv.org
November 13, 2025 at 9:39 PM
"Solving a Million-Step LLM Task with Zero Errors", from the inimitable Elliot Meyerson (et al.)
Reposted
It's fun to see your startup in axios: www.axios.com/2025/11/13/m...
Exclusive: Maryland taps AI for housing and government benefits
Applying for government assistance or housing permits, as well as processing those services, can be difficult and time consuming.
www.axios.com
November 13, 2025 at 9:54 PM
It's fun to see your startup in axios: www.axios.com/2025/11/13/m...
Reposted
BLUE ORIGIN LAUNCHES NEW GLENN ROCKET ON COMPANY'S FIRST NASA-SCALE SCIENCE MISSION, CARRYING TWIN MARS SATELLITES TO SPACE
November 13, 2025 at 8:56 PM
BLUE ORIGIN LAUNCHES NEW GLENN ROCKET ON COMPANY'S FIRST NASA-SCALE SCIENCE MISSION, CARRYING TWIN MARS SATELLITES TO SPACE
Reposted
Gregory Bovino, the top Border Patrol official leading the charge on the Trump administration’s immigration crackdown in cities, and his officers are heading to Charlotte, according to a source familiar with the planning. That was news to city officials. https://cnn.it/49crpZT
November 13, 2025 at 7:00 PM
Gregory Bovino, the top Border Patrol official leading the charge on the Trump administration’s immigration crackdown in cities, and his officers are heading to Charlotte, according to a source familiar with the planning. That was news to city officials. https://cnn.it/49crpZT
This is pretty cool to see, actual commercially available autonomous vehicle going from curb to curb on the highway.
I remember being at neurips in 2017 and everyone was hyped about AVs. took longer than most people thought, but 8 years isn't bad
www.reddit.com/r/SelfDrivin...
I remember being at neurips in 2017 and everyone was hyped about AVs. took longer than most people thought, but 8 years isn't bad
www.reddit.com/r/SelfDrivin...
From the SelfDrivingCars community on Reddit: My First Freeway Trip in a Waymo
Explore this post and more from the SelfDrivingCars community
www.reddit.com
November 13, 2025 at 5:14 PM
This is pretty cool to see, actual commercially available autonomous vehicle going from curb to curb on the highway.
I remember being at neurips in 2017 and everyone was hyped about AVs. took longer than most people thought, but 8 years isn't bad
www.reddit.com/r/SelfDrivin...
I remember being at neurips in 2017 and everyone was hyped about AVs. took longer than most people thought, but 8 years isn't bad
www.reddit.com/r/SelfDrivin...
Reposted
I just want to recognize for a moment how amazing it is that this image can appear before us unbidden in the comfort of our own homes
The First Photograph of a Human Being: A Photo Taken by Louis Daguerre in 1838
You’ve likely heard the reason people never smile in very old photographs. Early photography could be an excruciatingly slow process.
www.openculture.com
November 12, 2025 at 5:03 PM
I just want to recognize for a moment how amazing it is that this image can appear before us unbidden in the comfort of our own homes
Reposted
Miranda Priestly’s not riding the elevator alone anymore. The teaser trailer for “The Devil Wears Prada 2” has officially dropped.
Meryl Streep reunites with Anne Hathaway in ‘The Devil Wears Prada 2’ teaser: ‘Took you long enough’ | CNN
Miranda Priestly’s not riding the elevator alone anymore.
www.cnn.com
November 13, 2025 at 6:00 AM
Miranda Priestly’s not riding the elevator alone anymore. The teaser trailer for “The Devil Wears Prada 2” has officially dropped.
Reposted
Research shows local language models handle 88.7% of queries, cutting dependence on cloud infrastructure. With a 5.3x boost in energy efficiency since 2023, local AI transforms large-scale processing and conserves resources. https://arxiv.org/abs/2511.07885
Intelligence per Watt: Measuring Intelligence Efficiency of Local AI
ArXiv link for Intelligence per Watt: Measuring Intelligence Efficiency of Local AI
arxiv.org
November 13, 2025 at 12:21 AM
Research shows local language models handle 88.7% of queries, cutting dependence on cloud infrastructure. With a 5.3x boost in energy efficiency since 2023, local AI transforms large-scale processing and conserves resources. https://arxiv.org/abs/2511.07885
Reposted
Internal causal mechanisms in language models enhance predictions of out-of-distribution behavior, outperforming standard methods. With counterfactual simulation and value probing, this study increases model reliability and safety in practical applications. https://arxiv.org/abs/2505.11770
Internal Causal Mechanisms Robustly Predict Language Model Out-of-Distribution Behaviors
ArXiv link for Internal Causal Mechanisms Robustly Predict Language Model Out-of-Distribution Behaviors
arxiv.org
November 13, 2025 at 2:41 AM
Internal causal mechanisms in language models enhance predictions of out-of-distribution behavior, outperforming standard methods. With counterfactual simulation and value probing, this study increases model reliability and safety in practical applications. https://arxiv.org/abs/2505.11770
Reposted
New parallel algorithms accelerate sampling from complex distributions, achieving expected runtimes of eO(n^1/2) for autoregressive and diffusion models, using a novel "autospeculation" technique, enhancing efficiency in machine learning. https://arxiv.org/abs/2511.07869
Parallel Sampling via Autospeculation
ArXiv link for Parallel Sampling via Autospeculation
arxiv.org
November 13, 2025 at 3:11 AM
New parallel algorithms accelerate sampling from complex distributions, achieving expected runtimes of eO(n^1/2) for autoregressive and diffusion models, using a novel "autospeculation" technique, enhancing efficiency in machine learning. https://arxiv.org/abs/2511.07869
Pretty amazing result here. The energy efficiency of an electric vehicle is so good, if you have a truck towing an EV with regenerative braking until you use 1 gallon of gas, the EV charged regeneratively will go farther than the truck alone using 1 gallon of gas
www.facebook.com/share/v/1HoW...
www.facebook.com/share/v/1HoW...
Redirecting...
www.facebook.com
November 13, 2025 at 3:00 AM
Pretty amazing result here. The energy efficiency of an electric vehicle is so good, if you have a truck towing an EV with regenerative braking until you use 1 gallon of gas, the EV charged regeneratively will go farther than the truck alone using 1 gallon of gas
www.facebook.com/share/v/1HoW...
www.facebook.com/share/v/1HoW...
Reposted
Yeah man, "interest" isn't a positive if it's all conspiracy theories.
"Will the MeerKAT detection be enough to put an end to conspiracy theories about 3I/Atlas? Likely not, and at least the debates have heightened the general public's awareness of and interest in astrophysics."
"Will the MeerKAT detection be enough to put an end to conspiracy theories about 3I/Atlas? Likely not, and at least the debates have heightened the general public's awareness of and interest in astrophysics."
The First Radio Signal From Comet 3I/Atlas Ends the Debate About Its Nature
An observatory detected the first radio signal from the interstellar object 3I/Atlas. Here’s what it means.
www.wired.com
November 13, 2025 at 12:12 AM
Yeah man, "interest" isn't a positive if it's all conspiracy theories.
"Will the MeerKAT detection be enough to put an end to conspiracy theories about 3I/Atlas? Likely not, and at least the debates have heightened the general public's awareness of and interest in astrophysics."
"Will the MeerKAT detection be enough to put an end to conspiracy theories about 3I/Atlas? Likely not, and at least the debates have heightened the general public's awareness of and interest in astrophysics."
Reposted
AOC: "It is unconscionable that what we are debating right now is legislation that will give 8 senators over $1 million a piece & we're robbing people of their food assistance & healthcare to pay for it. How can we vote to enrich ourselves by stealing from the American people?"
November 12, 2025 at 10:39 PM
AOC: "It is unconscionable that what we are debating right now is legislation that will give 8 senators over $1 million a piece & we're robbing people of their food assistance & healthcare to pay for it. How can we vote to enrich ourselves by stealing from the American people?"
Apparently there’s a company fine tuning an ai model to generate an SVG and it does a great job
neosvg.com
neosvg.com
AI Text to SVG Generator | neoSVG
Generate perfect SVGs from text prompts using powerful AI.
neosvg.com
November 12, 2025 at 11:59 PM
Apparently there’s a company fine tuning an ai model to generate an SVG and it does a great job
neosvg.com
neosvg.com
I just left a comment in code specifically for an ai code reviewer to make sure a piece of debug code wasn’t enabled when checked in
A human reviewer would likely miss it but ai typically wouldn’t
A human reviewer would likely miss it but ai typically wouldn’t
November 12, 2025 at 9:43 PM
I just left a comment in code specifically for an ai code reviewer to make sure a piece of debug code wasn’t enabled when checked in
A human reviewer would likely miss it but ai typically wouldn’t
A human reviewer would likely miss it but ai typically wouldn’t
We’re past the uncanny valley in basically every medium for AI
Are you listening to bots? Survey shows AI music is virtually undetectable reut.rs/47FxooQ
Are you listening to bots? Survey shows AI music is virtually undetectable
A staggering 97% of listeners cannot distinguish between artificial intelligence-generated and human-composed songs, a Deezer–Ipsos survey showed on Wednesday, underscoring growing concerns that AI could upend how music is created, consumed and monetized.
reut.rs
November 12, 2025 at 9:31 PM
We’re past the uncanny valley in basically every medium for AI
Reposted
imo yes they are, afaict
when i’ve messed around with control vectors (activation manipulation), i couldn’t find a compelling use case. it all felt like prompting gave equivalent results
and if that’s the case, why should continual learning *literally* modify the model? that’s over-complicated..
when i’ve messed around with control vectors (activation manipulation), i couldn’t find a compelling use case. it all felt like prompting gave equivalent results
and if that’s the case, why should continual learning *literally* modify the model? that’s over-complicated..
Are prompting and activation steering just two sides of the same coin?
The paper formalizes a Bayesian framework for model control: altering a model's "beliefs" over which persona or data source it's emulating. Context (prompting) and internal representations (steering)
The paper formalizes a Bayesian framework for model control: altering a model's "beliefs" over which persona or data source it's emulating. Context (prompting) and internal representations (steering)
November 12, 2025 at 6:40 PM
imo yes they are, afaict
when i’ve messed around with control vectors (activation manipulation), i couldn’t find a compelling use case. it all felt like prompting gave equivalent results
and if that’s the case, why should continual learning *literally* modify the model? that’s over-complicated..
when i’ve messed around with control vectors (activation manipulation), i couldn’t find a compelling use case. it all felt like prompting gave equivalent results
and if that’s the case, why should continual learning *literally* modify the model? that’s over-complicated..
Systems safety is one of the things that keeps me up at night
Seemingly innocuous practices could actually be critically important as part of a bigger system. If too many people see their duties as trivial the entire thing breaks down
Applies to governments and engineering
Seemingly innocuous practices could actually be critically important as part of a bigger system. If too many people see their duties as trivial the entire thing breaks down
Applies to governments and engineering
The Most Dangerous Escalator in Europe
YouTube video by Veritasium
youtu.be
November 12, 2025 at 8:49 PM
Systems safety is one of the things that keeps me up at night
Seemingly innocuous practices could actually be critically important as part of a bigger system. If too many people see their duties as trivial the entire thing breaks down
Applies to governments and engineering
Seemingly innocuous practices could actually be critically important as part of a bigger system. If too many people see their duties as trivial the entire thing breaks down
Applies to governments and engineering
Reposted
Mass production of the new UBTECH walker s2 industrial humanoid robot
November 12, 2025 at 3:58 PM
Mass production of the new UBTECH walker s2 industrial humanoid robot
Reposted
This is really fun - here are my notes on how to try it out using Docker for Mac simonwillison.net/2025/Nov/12/...
November 12, 2025 at 5:17 PM
This is really fun - here are my notes on how to try it out using Docker for Mac simonwillison.net/2025/Nov/12/...
Hot take: skeuomorphism was never that bad and might actually be good
November 12, 2025 at 5:14 PM
Hot take: skeuomorphism was never that bad and might actually be good
You can literally see the frontier of human knowledge
AELLA is an open-science initiative to make scientific research accessible via structured summaries created by LLMs
Available now:
- Dataset of 100K summaries
- 2 fine-tuned LLMs
- 3d visualizer
Available now:
- Dataset of 100K summaries
- 2 fine-tuned LLMs
- 3d visualizer
November 12, 2025 at 5:36 AM
You can literally see the frontier of human knowledge
Reposted
AELLA is an open-science initiative to make scientific research accessible via structured summaries created by LLMs
Available now:
- Dataset of 100K summaries
- 2 fine-tuned LLMs
- 3d visualizer
Available now:
- Dataset of 100K summaries
- 2 fine-tuned LLMs
- 3d visualizer
November 12, 2025 at 5:35 AM
AELLA is an open-science initiative to make scientific research accessible via structured summaries created by LLMs
Available now:
- Dataset of 100K summaries
- 2 fine-tuned LLMs
- 3d visualizer
Available now:
- Dataset of 100K summaries
- 2 fine-tuned LLMs
- 3d visualizer