A.V.
@slckl.bsky.social
Trying to make Rust x AI a reality.
Python survivor, book lover and weird music enjoyer.
Python survivor, book lover and weird music enjoyer.
Reposted by A.V.
February 5, 2026 at 6:03 PM
Reposted by A.V.
Here’s one that’s not going to happen
February 4, 2026 at 8:40 PM
Here’s one that’s not going to happen
Reposted by A.V.
New CATL sodium ion batteries have:
- better performance in cold temps
- cheaper to make than lithium ion batteries
- significantly more stable and safer from fires.
- better performance in cold temps
- cheaper to make than lithium ion batteries
- significantly more stable and safer from fires.
January 27, 2026 at 12:53 AM
New CATL sodium ion batteries have:
- better performance in cold temps
- cheaper to make than lithium ion batteries
- significantly more stable and safer from fires.
- better performance in cold temps
- cheaper to make than lithium ion batteries
- significantly more stable and safer from fires.
A very sane AI usage policy for any open source project that still cares about quality.
AI Usage Policy
https://github.com/ghostty-org/ghostty/blob/main/AI_POLICY.md
https://news.ycombinator.com/item?id=46730504
https://github.com/ghostty-org/ghostty/blob/main/AI_POLICY.md
https://news.ycombinator.com/item?id=46730504
ghostty/AI_POLICY.md at main · ghostty-org/ghostty
👻 Ghostty is a fast, feature-rich, and cross-platform terminal emulator that uses platform-native UI and GPU acceleration. - ghostty-org/ghostty
github.com
January 23, 2026 at 6:10 PM
A very sane AI usage policy for any open source project that still cares about quality.
Reposted by A.V.
Democracy basically means electing a president. But the president,
January 21, 2026 at 3:06 PM
Democracy basically means electing a president. But the president,
A more efficient and more interpretable alternative to fat FFNs in Transformers. Sounds interesting...
Meta replaces FFN up-projection with a layer-local embedding lookup while keeping the gate and down-projection dense, enabling stable training, lower per-token compute, better interpretability, scalable parametric memory, and consistent accuracy gains without routing or communication overhead.
January 21, 2026 at 3:15 PM
A more efficient and more interpretable alternative to fat FFNs in Transformers. Sounds interesting...
New king of 30B released?
A model size that remains largely feasible for local deployments.
A model size that remains largely feasible for local deployments.
Zhipu just released a powerful lightweight option of GLM 4.7
✨ 30B total/3B active - MoE
huggingface.co/zai-org/GLM-...
✨ 30B total/3B active - MoE
huggingface.co/zai-org/GLM-...
January 19, 2026 at 7:25 PM
New king of 30B released?
A model size that remains largely feasible for local deployments.
A model size that remains largely feasible for local deployments.
Reposted by A.V.
DeepSeek’s new work: Engram 🔥
Beyond MoE, it adds lookup style conditional memory to LLMs.
Paper: github.com/deepseek-ai/...
Can’t wait to see what’s coming next 👀
Beyond MoE, it adds lookup style conditional memory to LLMs.
Paper: github.com/deepseek-ai/...
Can’t wait to see what’s coming next 👀
January 12, 2026 at 5:23 PM
DeepSeek’s new work: Engram 🔥
Beyond MoE, it adds lookup style conditional memory to LLMs.
Paper: github.com/deepseek-ai/...
Can’t wait to see what’s coming next 👀
Beyond MoE, it adds lookup style conditional memory to LLMs.
Paper: github.com/deepseek-ai/...
Can’t wait to see what’s coming next 👀
Reposted by A.V.
How many toddler sized robots do you think you could take in a fight
December 31, 2025 at 3:35 PM
How many toddler sized robots do you think you could take in a fight
Reposted by A.V.
Autonomous RIVR delivery robots in Pittsburgh
December 24, 2025 at 6:56 PM
Autonomous RIVR delivery robots in Pittsburgh
FoundationStereo was a meaningful boost for getting nice 3d results for folks like me who barely know what a point cloud is. This new version looks almost as good, but promises to be way faster. Fingers crossed for a friendly license 🤞
FastFoundationStereo from nvidia. Exciting because 3d information remains one of the easiest ways to get reliability and generalization; if this becomes practical, it can accelerate robot deployment quite a lot over pure RGB-based methods. github.com/NVlabs/Fast-...
December 17, 2025 at 8:45 PM
FoundationStereo was a meaningful boost for getting nice 3d results for folks like me who barely know what a point cloud is. This new version looks almost as good, but promises to be way faster. Fingers crossed for a friendly license 🤞
Reposted by A.V.
Nemotron 3
A new hybrid mamba2/attention LLM from NVIDIA that beats Qwen3-30B-A3B (same size & shape)
Notes:
* 1M context, with incredible recall past 256K
* New open datasets
* 10 open source RL environments
Overall this is a huge win for neolabs
huggingface.co/nvidia/NVIDI...
A new hybrid mamba2/attention LLM from NVIDIA that beats Qwen3-30B-A3B (same size & shape)
Notes:
* 1M context, with incredible recall past 256K
* New open datasets
* 10 open source RL environments
Overall this is a huge win for neolabs
huggingface.co/nvidia/NVIDI...
December 16, 2025 at 1:15 PM
Nemotron 3
A new hybrid mamba2/attention LLM from NVIDIA that beats Qwen3-30B-A3B (same size & shape)
Notes:
* 1M context, with incredible recall past 256K
* New open datasets
* 10 open source RL environments
Overall this is a huge win for neolabs
huggingface.co/nvidia/NVIDI...
A new hybrid mamba2/attention LLM from NVIDIA that beats Qwen3-30B-A3B (same size & shape)
Notes:
* 1M context, with incredible recall past 256K
* New open datasets
* 10 open source RL environments
Overall this is a huge win for neolabs
huggingface.co/nvidia/NVIDI...
Reposted by A.V.
Reposted by A.V.
Introducing Bolmo, a new family of byte-level language models built by "byteifying" our open Olmo 3—and to our knowledge, the first fully open byte-level LM to match or surpass SOTA subword models across a wide range of tasks. 🧵
December 15, 2025 at 5:19 PM
Introducing Bolmo, a new family of byte-level language models built by "byteifying" our open Olmo 3—and to our knowledge, the first fully open byte-level LM to match or surpass SOTA subword models across a wide range of tasks. 🧵
Reposted by A.V.
Here's a great review of what we saw in AI this year, from @gleech.org
AI in 2025: gestalt — LessWrong
This is the editorial for this year’s "Shallow Review of AI Safety". (It got long enough to stand alone.) …
www.lesswrong.com
December 8, 2025 at 5:24 PM
Here's a great review of what we saw in AI this year, from @gleech.org
Reposted by A.V.
Four new models from Mistral today - all Apache 2 licensed, all vision-capable, and one of them is a 3GB model that can run in a web browser and answer questions about things it can see through the webcam! simonwillison.net/2025/Dec/2/i...
Introducing Mistral 3
Four new models from Mistral today: three in their "Ministral" smaller model series (14B, 8B, and 3B) and a new Mistral Large 3 MoE model with 675B parameters, 41B active. …
simonwillison.net
December 2, 2025 at 5:32 PM
Four new models from Mistral today - all Apache 2 licensed, all vision-capable, and one of them is a 3GB model that can run in a web browser and answer questions about things it can see through the webcam! simonwillison.net/2025/Dec/2/i...
In other European news, kyutai labs, a non-profit ai research lab, spawned their (first?) for-profit branch: gradium.ai
With a 70M$ seed round, they look serious.
In their own words:
gradium.ai/blog/gradium
On the bad site, they even got a little promo video: x.com/GradiumAI/st...
With a 70M$ seed round, they look serious.
In their own words:
gradium.ai/blog/gradium
On the bad site, they even got a little promo video: x.com/GradiumAI/st...
Gradium
Text-to-Speech, Speech-to-Text, and Speech-to-Speech AI models
gradium.ai
December 2, 2025 at 7:28 PM
In other European news, kyutai labs, a non-profit ai research lab, spawned their (first?) for-profit branch: gradium.ai
With a 70M$ seed round, they look serious.
In their own words:
gradium.ai/blog/gradium
On the bad site, they even got a little promo video: x.com/GradiumAI/st...
With a 70M$ seed round, they look serious.
In their own words:
gradium.ai/blog/gradium
On the bad site, they even got a little promo video: x.com/GradiumAI/st...
Mistral dropped ministral 3B, 8B, 14B models and the big one - a seemingly deepseek shaped Mistral large 3, 675B moe brick. All apache 2!
Happy to see some European action in the usable model space.
Mistral blog post: mistral.ai/news/mistral-3
Happy to see some European action in the usable model space.
Mistral blog post: mistral.ai/news/mistral-3
December 2, 2025 at 7:19 PM
Mistral dropped ministral 3B, 8B, 14B models and the big one - a seemingly deepseek shaped Mistral large 3, 675B moe brick. All apache 2!
Happy to see some European action in the usable model space.
Mistral blog post: mistral.ai/news/mistral-3
Happy to see some European action in the usable model space.
Mistral blog post: mistral.ai/news/mistral-3
Reposted by A.V.
I’m running on a platform of Everyone Needs To Talk To Opus 4.5 For Two Hours
November 30, 2025 at 8:39 PM
I’m running on a platform of Everyone Needs To Talk To Opus 4.5 For Two Hours
Reposted by A.V.
At the risk of starting the flame war to end all flame wars...
Modern LLMs (GPT-5.1, Claude 4.5, Gemini 3) produce excellent code and can be a significant productivity boost to software engineers who take the time to learn how to effectively apply them - especially if used with coding agent tools
Modern LLMs (GPT-5.1, Claude 4.5, Gemini 3) produce excellent code and can be a significant productivity boost to software engineers who take the time to learn how to effectively apply them - especially if used with coding agent tools
November 27, 2025 at 7:55 PM
At the risk of starting the flame war to end all flame wars...
Modern LLMs (GPT-5.1, Claude 4.5, Gemini 3) produce excellent code and can be a significant productivity boost to software engineers who take the time to learn how to effectively apply them - especially if used with coding agent tools
Modern LLMs (GPT-5.1, Claude 4.5, Gemini 3) produce excellent code and can be a significant productivity boost to software engineers who take the time to learn how to effectively apply them - especially if used with coding agent tools
Reposted by A.V.
And a major open science release from Prime Intellect: they don’t stress it enough but SFT part is beyond post-training. This is a fully documented mid-training with tons of insights/gems on MoE training, asynchronous infra RL, deep research. storage.googleapis.com/intellect-3-...
November 27, 2025 at 7:47 AM
And a major open science release from Prime Intellect: they don’t stress it enough but SFT part is beyond post-training. This is a fully documented mid-training with tons of insights/gems on MoE training, asynchronous infra RL, deep research. storage.googleapis.com/intellect-3-...
SAM3 dropped for those who celebrate!
Similar to SAM2, it can segment stuff based on points and track stuff, but now it can directly segment stuff based on text and image prompts, too.
Webpage: ai.meta.com/sam3/
Repo: github.com/facebookrese...
Similar to SAM2, it can segment stuff based on points and track stuff, but now it can directly segment stuff based on text and image prompts, too.
Webpage: ai.meta.com/sam3/
Repo: github.com/facebookrese...
SAM 3
With SAM 3 you can use text and visual prompts to precisely identify, segment, and follow any object in images or videos—coming soon to Instagram Edits and Vibes on the Meta AI app.
ai.meta.com
November 19, 2025 at 6:16 PM
SAM3 dropped for those who celebrate!
Similar to SAM2, it can segment stuff based on points and track stuff, but now it can directly segment stuff based on text and image prompts, too.
Webpage: ai.meta.com/sam3/
Repo: github.com/facebookrese...
Similar to SAM2, it can segment stuff based on points and track stuff, but now it can directly segment stuff based on text and image prompts, too.
Webpage: ai.meta.com/sam3/
Repo: github.com/facebookrese...
Reposted by A.V.
Me: I want to have more friends
Tech companies:
Tech companies:
November 19, 2025 at 3:51 PM
Me: I want to have more friends
Tech companies:
Tech companies:
I love widely used high quality datasets.
November 18, 2025 at 8:14 AM
I love widely used high quality datasets.