Manu Manu
@manumanu.bsky.social
Lead Prog, interested in most tech subjects, working at Ubisoft Paris
Reposted by Manu Manu
"Hey, Gemini 3, So I need DOOM, but more root vegetables, also no guns or demons or mars. And more of a focus on different flooring styles. but otherwise EXACTLY the same as DOOM."
Gemini: "Here is F.L.O.O.R. (First-person Lino Observation & Ornamental Review)."
Pretty good!
Gemini: "Here is F.L.O.O.R. (First-person Lino Observation & Ornamental Review)."
Pretty good!
November 19, 2025 at 9:08 PM
"Hey, Gemini 3, So I need DOOM, but more root vegetables, also no guns or demons or mars. And more of a focus on different flooring styles. but otherwise EXACTLY the same as DOOM."
Gemini: "Here is F.L.O.O.R. (First-person Lino Observation & Ornamental Review)."
Pretty good!
Gemini: "Here is F.L.O.O.R. (First-person Lino Observation & Ornamental Review)."
Pretty good!
Reposted by Manu Manu
Meta released a new generation of Segment Anything Models:
- SAM 3 enables detecting, segmenting and tracking of objects across images and videos, now with short text phrases and exemplar prompts.
ai.meta.com/blog/segment...
- SAM 3 enables detecting, segmenting and tracking of objects across images and videos, now with short text phrases and exemplar prompts.
ai.meta.com/blog/segment...
Introducing Meta Segment Anything Model 3 and Segment Anything Playground
Explore Segment Anything Model 3 and the new Segment Anything Playground, a place to experience the full capabilities of our most advanced SAM releases to date.
ai.meta.com
November 20, 2025 at 3:36 AM
Meta released a new generation of Segment Anything Models:
- SAM 3 enables detecting, segmenting and tracking of objects across images and videos, now with short text phrases and exemplar prompts.
ai.meta.com/blog/segment...
- SAM 3 enables detecting, segmenting and tracking of objects across images and videos, now with short text phrases and exemplar prompts.
ai.meta.com/blog/segment...
Reposted by Manu Manu
Stereogum says Google's switch to AI Overviews reduced its ad revenue by *70 percent* www.theverge.com/entertainmen...
Stereogum soldiers on in the era of streaming and AI
Stereogum Founder Scott Lapatine on the importance of independent media in the age of AI slop and algorithms.
www.theverge.com
November 17, 2025 at 4:26 PM
Stereogum says Google's switch to AI Overviews reduced its ad revenue by *70 percent* www.theverge.com/entertainmen...
Reposted by Manu Manu
I’ve had a running test for multimodal models over the last ~year: extract the “Crss” values for each tick mark in this chart. GPT 5 makes up numbers, Claude 4.5 refuses to try, Qwen3 VL has a plausible guess. Gemini 3 pro pretty much nails it.
November 19, 2025 at 1:00 AM
I’ve had a running test for multimodal models over the last ~year: extract the “Crss” values for each tick mark in this chart. GPT 5 makes up numbers, Claude 4.5 refuses to try, Qwen3 VL has a plausible guess. Gemini 3 pro pretty much nails it.
Reposted by Manu Manu
Just image Transformers (JiT)
JiTs are simple large-patch Transformers that operate on raw pixels, no tokenizer, pre-training, or extra losses needed. JiT excels in high-dimensional spaces where traditional noise-predicting models can fail.
JiTs are simple large-patch Transformers that operate on raw pixels, no tokenizer, pre-training, or extra losses needed. JiT excels in high-dimensional spaces where traditional noise-predicting models can fail.
November 19, 2025 at 5:42 AM
Just image Transformers (JiT)
JiTs are simple large-patch Transformers that operate on raw pixels, no tokenizer, pre-training, or extra losses needed. JiT excels in high-dimensional spaces where traditional noise-predicting models can fail.
JiTs are simple large-patch Transformers that operate on raw pixels, no tokenizer, pre-training, or extra losses needed. JiT excels in high-dimensional spaces where traditional noise-predicting models can fail.
Reposted by Manu Manu
Alibaba released AgentEvolver , an open-source, self-evolving agent system.
AgentEvolver integrates three synergistic mechanisms: Self-Questioning, Self-Navigating, and Self-Attributing; to systematically address critical bottlenecks in Agent RL training, including task scarcity,
AgentEvolver integrates three synergistic mechanisms: Self-Questioning, Self-Navigating, and Self-Attributing; to systematically address critical bottlenecks in Agent RL training, including task scarcity,
November 19, 2025 at 5:46 AM
Alibaba released AgentEvolver , an open-source, self-evolving agent system.
AgentEvolver integrates three synergistic mechanisms: Self-Questioning, Self-Navigating, and Self-Attributing; to systematically address critical bottlenecks in Agent RL training, including task scarcity,
AgentEvolver integrates three synergistic mechanisms: Self-Questioning, Self-Navigating, and Self-Attributing; to systematically address critical bottlenecks in Agent RL training, including task scarcity,
Reposted by Manu Manu
Today we’re releasing Deep Research Tulu (DR Tulu)—the first fully open, end-to-end recipe for long-form deep research, plus an 8B agent you can use right away. Train agents that plan, search, synthesize, & cite across sources, making expert research more accessible. 🧭📚
November 18, 2025 at 3:31 PM
Today we’re releasing Deep Research Tulu (DR Tulu)—the first fully open, end-to-end recipe for long-form deep research, plus an 8B agent you can use right away. Train agents that plan, search, synthesize, & cite across sources, making expert research more accessible. 🧭📚
Reposted by Manu Manu
I’m not sure if it’s worth looking at, but he did name his project in a way that’s guaranteed to go viral.
Karpathy
An agentic ML Engineer that trains state-of-the-art ML models using Claude Code SDK and Google ADK. This is a very simple implemenation demoing the power of Claude skills for ML.
Karpathy
An agentic ML Engineer that trains state-of-the-art ML models using Claude Code SDK and Google ADK. This is a very simple implemenation demoing the power of Claude skills for ML.
GitHub - K-Dense-AI/karpathy: An agentic Machine Learning Engineer
An agentic Machine Learning Engineer. Contribute to K-Dense-AI/karpathy development by creating an account on GitHub.
github.com
November 19, 2025 at 6:10 AM
I’m not sure if it’s worth looking at, but he did name his project in a way that’s guaranteed to go viral.
Karpathy
An agentic ML Engineer that trains state-of-the-art ML models using Claude Code SDK and Google ADK. This is a very simple implemenation demoing the power of Claude skills for ML.
Karpathy
An agentic ML Engineer that trains state-of-the-art ML models using Claude Code SDK and Google ADK. This is a very simple implemenation demoing the power of Claude skills for ML.
Reposted by Manu Manu
“Helsinki hasn’t registered a single traffic-related fatality in the past year…Citing data that shows the risk of pedestrian fatality is cut in half by reducing a car’s speed from 40 to 30km/hr, city officials imposed the lower limit in most of Helsinki’s residential areas & city center in 2021.”
Helsinki just went a full year without a single traffic death
The capital city is Finnish’ed with car-related fatalities.
www.politico.eu
November 17, 2025 at 6:31 AM
“Helsinki hasn’t registered a single traffic-related fatality in the past year…Citing data that shows the risk of pedestrian fatality is cut in half by reducing a car’s speed from 40 to 30km/hr, city officials imposed the lower limit in most of Helsinki’s residential areas & city center in 2021.”
Reposted by Manu Manu
Funny.
For more than one reason...
For more than one reason...
November 17, 2025 at 8:32 AM
Funny.
For more than one reason...
For more than one reason...
Another interesting World model !
Video models, world models... They all aim to give genAI better understanding of our world, where text is still limited.
ifm.mbzuai.ac.ae/pan/
Video models, world models... They all aim to give genAI better understanding of our world, where text is still limited.
ifm.mbzuai.ac.ae/pan/
PAN – Interactive World Model by MBZUAI
From the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), meet PAN — an interactive world model that can see, reason, and predict how the world changes over time.
ifm.mbzuai.ac.ae
November 16, 2025 at 3:22 PM
Another interesting World model !
Video models, world models... They all aim to give genAI better understanding of our world, where text is still limited.
ifm.mbzuai.ac.ae/pan/
Video models, world models... They all aim to give genAI better understanding of our world, where text is still limited.
ifm.mbzuai.ac.ae/pan/
Reposted by Manu Manu
It would have taken former xDefiant executive producer Mark Rubin less time to read the story than it did to write this long but inaccurate tweet. As the actual story says, the team that would go on to make Dispatch worked on Splinter Cell in 2017, two years before Rubin even joined Ubisoft.
November 15, 2025 at 12:51 PM
It would have taken former xDefiant executive producer Mark Rubin less time to read the story than it did to write this long but inaccurate tweet. As the actual story says, the team that would go on to make Dispatch worked on Splinter Cell in 2017, two years before Rubin even joined Ubisoft.
Reposted by Manu Manu
New: Google has chosen a side in Trump's mass deportation campaign. Google is hosting a CBP facial recognition app to hunt immigrants; no indication Google will remove. At same time Google takes down apps for reporting ICE sightings
“Big tech has made their choice”
www.404media.co/google-has-c...
“Big tech has made their choice”
www.404media.co/google-has-c...
Google Has Chosen a Side in Trump's Mass Deportation Effort
Google is hosting a CBP app that uses facial recognition to identify immigrants, while simultaneously removing apps that report the location of ICE officials because Google sees ICE as a vulnerable gr...
www.404media.co
November 13, 2025 at 2:11 PM
New: Google has chosen a side in Trump's mass deportation campaign. Google is hosting a CBP facial recognition app to hunt immigrants; no indication Google will remove. At same time Google takes down apps for reporting ICE sightings
“Big tech has made their choice”
www.404media.co/google-has-c...
“Big tech has made their choice”
www.404media.co/google-has-c...
Reposted by Manu Manu
if you tax billionaires and trillionaires you'll generate an absolute shitton of new millionaires
NEW: The first two years of Massachusetts' millionaire tax has raised $3 billion more than expected.
And rather than driving the rich away, IPS researchers found that the number of millionaires has *increased.*
Tax the rich. Greg Ryan in @bloomberg.com:
And rather than driving the rich away, IPS researchers found that the number of millionaires has *increased.*
Tax the rich. Greg Ryan in @bloomberg.com:
Millionaire Tax That Inspired Mamdani Fuels $5.7 Billion Haul in Massachusetts
A millionaire levy in Massachusetts that New York City mayoral frontrunner Zohran Mamdani holds up as a model for taxing the rich has generated $3 billion more in revenue than expected without forcing...
www.bloomberg.com
November 14, 2025 at 2:53 AM
if you tax billionaires and trillionaires you'll generate an absolute shitton of new millionaires
Reposted by Manu Manu
Only available in China? Doubao-Seed-Code model is not one of the available model for Trae ( docs.trae.ai/ide/models?_... )
ByteDance unveils China’s most affordable AI coding agent at just US$1.30 a month
www.scmp.com/tech/big-tec...
ByteDance unveils China’s most affordable AI coding agent at just US$1.30 a month
www.scmp.com/tech/big-tec...
ByteDance unveils China’s most affordable AI coding agent at just US$1.30 a month
The TikTok owner’s cloud unit has launched a new AI coding assistant, stoking China’s AI coding price war.
www.scmp.com
November 13, 2025 at 6:48 AM
Only available in China? Doubao-Seed-Code model is not one of the available model for Trae ( docs.trae.ai/ide/models?_... )
ByteDance unveils China’s most affordable AI coding agent at just US$1.30 a month
www.scmp.com/tech/big-tec...
ByteDance unveils China’s most affordable AI coding agent at just US$1.30 a month
www.scmp.com/tech/big-tec...
Reposted by Manu Manu
MUVERA
Multi-Vector Retrieval via Fixed Dimensional Encodings is an interesting approach by Google Research. It transforms multi-vector representations into single fixed-size vectors (fixed dimensional encodings).
Multi-Vector Retrieval via Fixed Dimensional Encodings is an interesting approach by Google Research. It transforms multi-vector representations into single fixed-size vectors (fixed dimensional encodings).
November 12, 2025 at 5:36 AM
MUVERA
Multi-Vector Retrieval via Fixed Dimensional Encodings is an interesting approach by Google Research. It transforms multi-vector representations into single fixed-size vectors (fixed dimensional encodings).
Multi-Vector Retrieval via Fixed Dimensional Encodings is an interesting approach by Google Research. It transforms multi-vector representations into single fixed-size vectors (fixed dimensional encodings).
Reposted by Manu Manu
So it turns out... the US air travel system was incredibly, deeply dependent on federal funding to just run day-to-day all this time, to the benefit of private airline shareholders, when everyone thinks that state-run trains are leeching off the government. Weird!
November 9, 2025 at 12:10 AM
So it turns out... the US air travel system was incredibly, deeply dependent on federal funding to just run day-to-day all this time, to the benefit of private airline shareholders, when everyone thinks that state-run trains are leeching off the government. Weird!
The ShowReel 2025 from Godot is nice !
Godot games really improved in quality and variety !
www.youtube.com/watch?v=7ZwE...
Godot games really improved in quality and variety !
www.youtube.com/watch?v=7ZwE...
Godot Engine – 2025 Showreel
YouTube video by Godot Engine
www.youtube.com
November 8, 2025 at 11:45 PM
The ShowReel 2025 from Godot is nice !
Godot games really improved in quality and variety !
www.youtube.com/watch?v=7ZwE...
Godot games really improved in quality and variety !
www.youtube.com/watch?v=7ZwE...
Reposted by Manu Manu
Apple will pay Google about $1 billion annually to use Google’s AI model for Siri, but Perplexity will pay Snap about $400 million so Snapchat will use its AI.
I don't know..... this doesn’t exactly sound like a good deal for Perplexity.
I don't know..... this doesn’t exactly sound like a good deal for Perplexity.
November 6, 2025 at 12:25 PM
Apple will pay Google about $1 billion annually to use Google’s AI model for Siri, but Perplexity will pay Snap about $400 million so Snapchat will use its AI.
I don't know..... this doesn’t exactly sound like a good deal for Perplexity.
I don't know..... this doesn’t exactly sound like a good deal for Perplexity.
Reposted by Manu Manu
I joked recently at work that another game targeting May 26th could have been dumb-genius move as GTA could get delayed again and, well, here it is.
BREAKING: GTA 6 has been delayed (again) until November 2026, Rockstar confirms
Grand Theft Auto 6 Delayed Again Until November 2026 - Kotaku
The highly anticipated open world game was set to launch in May 2026
kotaku.com
November 6, 2025 at 10:29 PM
I joked recently at work that another game targeting May 26th could have been dumb-genius move as GTA could get delayed again and, well, here it is.
Reposted by Manu Manu
New blog post! Behind the scenes of some of the techniques involved in making our last PC demo 💫 gboisse.github.io/posts/this-i...
November 5, 2025 at 9:15 PM
New blog post! Behind the scenes of some of the techniques involved in making our last PC demo 💫 gboisse.github.io/posts/this-i...
Reposted by Manu Manu
DeltaNet Explained by Sonlin Yang
A gentle and comprehensive introduction to the DeltaNet
Part 1: sustcsonglin.github.io/blog/2024/de...
Part 2: sustcsonglin.github.io/blog/2024/de...
Part 3: sustcsonglin.github.io/blog/2024/de...
A gentle and comprehensive introduction to the DeltaNet
Part 1: sustcsonglin.github.io/blog/2024/de...
Part 2: sustcsonglin.github.io/blog/2024/de...
Part 3: sustcsonglin.github.io/blog/2024/de...
November 5, 2025 at 11:45 PM
DeltaNet Explained by Sonlin Yang
A gentle and comprehensive introduction to the DeltaNet
Part 1: sustcsonglin.github.io/blog/2024/de...
Part 2: sustcsonglin.github.io/blog/2024/de...
Part 3: sustcsonglin.github.io/blog/2024/de...
A gentle and comprehensive introduction to the DeltaNet
Part 1: sustcsonglin.github.io/blog/2024/de...
Part 2: sustcsonglin.github.io/blog/2024/de...
Part 3: sustcsonglin.github.io/blog/2024/de...