TechTalks
Subscribe
Sign in
Home
Archive
About
Inside Nvidia's new technique to optimize long-context inference and continual learning
By treating language modeling as a continual learning problem, the TTT-E2E architecture achieves the accuracy of full-attention Transformers on 128k…
Jan 13
•
Ben Dickson
3
Meta’s new VL-JEPA model shifts from generating tokens to predicting concepts
Meta’s VL-JEPA outperforms massive vision-language models on world modeling tasks by learning to predict "thought vectors" instead of text tokens.
Jan 4
•
Ben Dickson
9
6
4
How reinforcement learning changed LLM tool-use
A look at the evolution of LLM tool-use, from supervised fine-tuning to Reinforcement Learning (RLVR) and agentic applications in large and specialized…
Dec 30, 2025
•
Ben Dickson
8
2
Inside URM, the architecture beating standard Transformers on reasoning tasks
The key to solving complex reasoning isn't stacking more transformer layers, but refining the "thought process" through efficient recurrent loops.
Dec 24, 2025
•
Ben Dickson
8
4
What (I think) makes Gemini 3 Flash so good and fast
Google didn’t reveal a lot of information about its new Flash model. So we had to speculate a lot on what is going on under the hood.
Dec 22, 2025
•
Ben Dickson
10
5
1
Latest
Top
Discussions
Nvidia sets a new bar for open source models with Nemotron 3
As the industry shifts from chatbots to multi-agent workflows, Nvidia's Nemotron 3 offers a blueprint for efficient, long-context reasoning.
Dec 17, 2025
•
Ben Dickson
5
4
2
AI benchmarks are confusing. Here's why.
AI labs are racing to overtake each other on key industry benchmarks. But this intense race has stripped the benchmarks of most of their value.
Dec 16, 2025
•
Ben Dickson
1
2
Salesforce takes a new approach to web agents with WALT
The framework abstracts away the chaos of dynamic layouts, allowing AI to focus on high-level planning and tool-use instead of low-level clicks.
Dec 12, 2025
•
Ben Dickson
3
2
Poetiq crushed ARC-AGI-2 at half the cost of Gemini 3 Deep Think. Here's how.
The verified solution achieves 54% accuracy on the semi-private test set, outperforming Gemini 3 Deep Think at less than half the cost.
Dec 10, 2025
•
Ben Dickson
14
1
4
The magic sauce that makes DeepSeek-V3.2 so damn efficient
DeepSeek-V3.2 is a top-5 LLM, sitting next to the likes of Grok 4 and GPT-5. But what is more impressive is its cost-efficiency and magnificent…
Dec 6, 2025
•
Ben Dickson
11
3
What OpenAI's 'Code Red' says about the current state of the AI race
OpenAI’s problem is not that it doesn't have the best model anymore but that the general feeling is that it has fallen behind.
Dec 4, 2025
•
Ben Dickson
9
1
1
What comes for LLMs after reinforcement learning with verifiable rewards (RLVR)?
Reinforcement learning from verifiable rewards (RLVR) ushered in a new generation of reasoning models. Now, researchers are looking beyond RLVR for the…
Dec 2, 2025
•
Ben Dickson
1
2
See all
TechTalks
In-depth discussions about machine learning, deep learning, reinforcement learning, neural networks, artificial general intelligence, AI business, and other technology trends.
Subscribe
Recommendations
View all 12
The Founders Corner®
Ruben Dominguez
Colligo
Erik J Larson
AI: A Guide for Thinking Humans
Melanie Mitchell
AI Supremacy
Michael Spencer
Sutskever's List
Rich Heimann
TechTalks
Subscribe
About
Archive
Recommendations
Sitemap
This site requires JavaScript to run correctly. Please
turn on JavaScript
or unblock scripts