DeepMind - TechTalks

What to know about Google Gemini 2.5 Pro

Ben Dickson — Wed, 26 Mar 2025 18:12:09 +0000

Gemini 2.5 Pro is a new reasoning model that excels in long-context tasks and benchmarks, revitalizing Google’s AI strategy against competitors like OpenAI.

The post What to know about Google Gemini 2.5 Pro first appeared on TechTalks.

How multiagent fine-tuning overcomes the data bottleneck of LLMs

Ben Dickson — Mon, 27 Jan 2025 17:03:28 +0000

Multiagent debate and fine-tuning can enable LLMs to create high-quality training data to improve themselves across different tasks.

The post How multiagent fine-tuning overcomes the data bottleneck of LLMs first appeared on TechTalks.

DeepMind releases benchmark for evaluating long-context LLMs

Ben Dickson — Mon, 01 Jul 2024 07:50:18 +0000

Google DeepMind has released Long-Context Frontiers (LOFT), a benchmark for LLMs that can process hundreds of thousands or millions of tokens in one prompt.

The post DeepMind releases benchmark for evaluating long-context LLMs first appeared on TechTalks.

Optimize your ChatGPT prompts with DeepMind’s OPRO technique

Ben Dickson — Mon, 20 Nov 2023 14:00:00 +0000

Optimization by PROmpting (OPRO), a powerful method developed by Google DeepMind, uses large language models (LLM) as optimizers for their own prompts.

The post Optimize your ChatGPT prompts with DeepMind’s OPRO technique first appeared on TechTalks.

LLMs can’t self-correct in reasoning tasks, DeepMind study finds

Ben Dickson — Mon, 09 Oct 2023 13:00:00 +0000

A study by Google's DeepMind and the University of Illinois at Urbana-Champaign has found that self-correction in large language models (LLMs) isn't universally effective.

The post LLMs can’t self-correct in reasoning tasks, DeepMind study finds first appeared on TechTalks.

DeepMind AlphaTensor: The delicate balance between human and artificial intelligence

Ben Dickson — Mon, 10 Oct 2022 13:00:00 +0000

DeepMind AlphaTensor shows how the right combination of human and artificial intelligence can find solutions to complicated problems.

The post DeepMind AlphaTensor: The delicate balance between human and artificial intelligence first appeared on TechTalks.

What DeepMind’s AlphaCode is and isn’t

Ben Dickson — Mon, 07 Feb 2022 14:00:00 +0000

This article is part of our reviews of AI research papers, a series of posts that explore the latest findings in artificial intelligence. DeepMind is the latest AI research lab to introduce a deep learning model that can generate software source code with remarkable results. Called AlphaCode, the model is based on Transformers, the same architecture […]

The post What DeepMind’s AlphaCode is and isn’t first appeared on TechTalks.

DeepMind RL method promises better co-op between AI and humans

Ben Dickson — Mon, 22 Nov 2021 14:00:00 +0000

AI researchers at DeepMind present a new technique to improve the capacity of reinforcement learning agents to cooperate with humans at different skill levels.

The post DeepMind RL method promises better co-op between AI and humans first appeared on TechTalks.

What MuJoCo’s acquisition means for DeepMind’s AI & robotics research

Ben Dickson — Fri, 22 Oct 2021 13:00:00 +0000

DeepMind has acquired and open-sourced physics simulator MuJoCo. It can have a positive impact on the robotics community and the future of the AI research lab.

The post What MuJoCo’s acquisition means for DeepMind’s AI & robotics research first appeared on TechTalks.

DeepMind becomes profitable—and more enmeshed in Google’s business

Ben Dickson — Thu, 07 Oct 2021 13:00:00 +0000

DeepMind has finally become profitable. But it is also becoming inextricably tied to its parent company Google.

The post DeepMind becomes profitable—and more enmeshed in Google’s business first appeared on TechTalks.

Is DeepMind’s new reinforcement learning system a step toward general AI?

Ben Dickson — Mon, 02 Aug 2021 13:00:00 +0000

DeepMind has released a new paper that shows impressive advances in reinforcement learning. How far does it bring us toward general AI?

The post Is DeepMind’s new reinforcement learning system a step toward general AI? first appeared on TechTalks.

Building artificial intelligence: Reward is not enough

Herbert Roitblat — Wed, 07 Jul 2021 13:00:00 +0000

Scientist and author Herbert Roitblat provides an alternate view on DeepMind's "Reward is enough" paper.

The post Building artificial intelligence: Reward is not enough first appeared on TechTalks.

Evolution, rewards, and artificial intelligence

Ben Dickson — Thu, 17 Jun 2021 13:00:00 +0000

A paper by DeepMind scientist triggered much debate about the path to artificial intelligence. Here, we'll try to draw the line between theory and practice.

The post Evolution, rewards, and artificial intelligence first appeared on TechTalks.

DeepMind scientists: Reinforcement learning is enough for general AI

Ben Dickson — Mon, 07 Jun 2021 12:30:00 +0000

In a new paper, scientists at DeepMind suggest that reward maximization and reinforcement learning are enough to develop artificial general intelligence.

The post DeepMind scientists: Reinforcement learning is enough for general AI first appeared on TechTalks.

DeepMind’s annual report: Why it’s hard to run a commercial AI lab

Ben Dickson — Mon, 21 Dec 2020 13:19:08 +0000

DeepMinds growing losses highlight the challenges of running an artificial intelligence research lab in the corporate world.

The post DeepMind’s annual report: Why it’s hard to run a commercial AI lab first appeared on TechTalks.

Neuroscience shows what’s right and wrong with AI

Ben Dickson — Mon, 20 Jan 2020 14:00:10 +0000

Work by neuroscientists and AI researchers at DeepMind and labs in Germany and Greece provides a clear view of what's working in artificial intelligence.

The post Neuroscience shows what’s right and wrong with AI first appeared on TechTalks.

DeepMind AlphaStar: AI breakthrough or pushing the limits of reinforcement learning?

Ben Dickson — Mon, 04 Nov 2019 14:00:58 +0000

DeepMind's AlphaStar proves we can still push the limits of AI. But it also reminds us of the challenges we must overcome to replicate the human brain.

The post DeepMind AlphaStar: AI breakthrough or pushing the limits of reinforcement learning? first appeared on TechTalks.

DeepMind’s inner feud: The problem with commercializing scientific AI research

Ben Dickson — Mon, 26 Aug 2019 13:00:23 +0000

The apparent infighting at AI research lab DeepMind shows why it's wrong to mingle commercial and scientific AI.

The post DeepMind’s inner feud: The problem with commercializing scientific AI research first appeared on TechTalks.

DeepMind’s losses display the challenges of the AI industry

Ben Dickson — Mon, 12 Aug 2019 13:00:19 +0000

DeepMind’s huge costs bring to light some of the most serious challenges the artificial intelligence industry is grappling with.

The post DeepMind’s losses display the challenges of the AI industry first appeared on TechTalks.

The interesting facts behind DeepMind’s Quake-playing AI

Ben Dickson — Mon, 03 Jun 2019 13:00:19 +0000

DeepMind latest paper introduced an AI that has mastered the game Quake III. While Quake isn't the most complicated game AI has mastered, but it still presents some interesting challenges.

The post The interesting facts behind DeepMind’s Quake-playing AI first appeared on TechTalks.