Follow the latest in AI
We’ve been building a brand-new kind of jury lately, and almost nobody has noticed. It doesn’t sit in a wooden box or complain about parking. It’s a panel of AI models, and its job is to deliver verdicts — Is this answer helpful? Is this comment toxic? Is this summary any good? We line up several AIs, … Read more
From sodium-ion grids to embryo scoring, MIT Technology Review’s 2026 list—presented at SXSW—maps where tech, energy, and biology are heading.
The AI industry has spent the last year obsessed with “Agents”—AI systems that can use tools, browse the web, and execute multi-step tasks. But the conversation is already shifting. We are moving past simple task-execution toward a far more radical horizon: Recursive Self-Improvement and Institutional Intelligence. Two seminal papers from March 2026, Hyperagents (Zhang et … Read more
In my last article, we explored how DeepSeek’s Engram effectively gave AI a hippocampus—offloading static facts into a massive, efficient lookup table. It was a breakthrough in separating memory from reasoning. But what if the model didn’t just “look up” memories? What if it actually rewired its own brain while it was reading, optimizing its … Read more
The race for Artificial General Intelligence (AGI) has hit a bottleneck: efficiency. Standard Large Language Models (LLMs) are “computationally heavy” because they don’t know how to separate thinking from remembering. A groundbreaking research paper from Peking University and DeepSeek-AI—“Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models” (arXiv:2601.07372)—is changing the … Read more
LLM based autonomous agents can use Generative AI to automate processes without needing human intervention. There is a subtle distinction between autonomous agents and workflows as explained in this anthropic blog. If you know the series of steps needed to automate a process, then you can use a workflow. However, if you need a system … Read more
When a data scientist starts a training job, it is useful to enforce tagging so that cost can be allocated to the training job. We need to enforce both the tag key and the value. For example, a tag key can be ‘project’ and its value could be ‘projectA’. The user should not be able … Read more
DeepSpeed library optimizes training and inference for foundational models. Learn more about its features.
CLIP introduces a model that enables zero shot learning for a new dataset (in addition to a new example) by using NLP to supervise pre-training. i.e., To identify an object, you can provide the name or description of a new object that the model has not seen before. Traditionally a computer vision model was trained … Read more
Why S3 lifecycle policies S3 lifecycle policies allow you to do two things: Reduce cost by deleting data that is not longer required. Implement your security policies by : Retaining data for the required duration and reducing cost by moving it to low cost storage. Deleting data that you are not allowed to retain for … Read more