TechTalks

TechTalks

Home
Archive
About
Key lessons from Anthropic's 'agentic misalignment' study
Anthropic's study warns that LLMs may intentionally act harmfully under pressure, foreshadowing the potential risks of agentic systems without human…
 • 
Ben Dickson
1
The illusion of 'thoughts' and 'reasoning' in LLMs
Chain-of-thought tokens don't reflect genuine reasoning in LLMs is misleading. They're navigational aids devoid of true cognitive processing or…
 • 
Ben Dickson
1
Why Anthropic’s bet against Windsurf is a mistake
Anthropic's abrupt decision to cut Windsurf could potentially pave the way for competitors like Google amid market evolution.
 • 
Ben Dickson
1
Why the factory is the first real market for humanoid robots
Humanoid robots may advance, but starting with home deployment faces safety concerns, unique challenges, and uncertain consumer demand. Factories make…
 • 
Ben Dickson

May 2025

The fascinating synergy between AI and neuroscience
A discussion with Dr. Rahul Biswas, postdoctoral scholar at UCSF, on how advances AI and neuroscience are feeding into each other.
 • 
Ben Dickson
2
LLMs reasoning traces can be misleading
There is significant doubt about the trustworthiness of chain-of-thought traces in large language models, challenging developers' reliance on them for…
 • 
Ben Dickson
1
The human bottleneck and accountability: AI's two critical hurdles in production
Klarna's reversal on AI demonstrates the pitfalls of overhyped technology. Human oversight and accountability remains crucial, even if it hinders AI…
 • 
Ben Dickson
1
An AI system that is more 'human-like'
Sakana AI's Continuous Thought Machine enhances AI's alignment with human cognition, promising a future of more trustworthy and efficient technology.
 • 
Ben Dickson
1
Why AI won't kill programming
Programming isn't dead; it's evolving. Vibe coding opens doors, yet true mastery demands deep skills. Embrace the change and thrive in this new tech…
 • 
Ben Dickson
2
New RL technique makes diffusion LLMs viable contenders for reasoning tasks
Diffusion LLMs outperform autoregressive models in simple inference. Now they're getting reasoning abilities.
 • 
Ben Dickson
Improving LLM reasoning with step-wise reinforcement learning
SWiRL trains models to interleave reasoning, tool-use, and answer generation, making it useful for agentic applications.
 • 
Ben Dickson
How effective are inference-time scaling techniques?
Simply throwing more compute at a problem doesn't guarantee better or more efficient results.
 • 
Ben Dickson
© 2025 Ben Dickson
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture