Digg - AI news, before it trends

Dwarkesh Patel separates intelligence from goal achievement

Dwarkesh Patel argues that intelligence should be viewed as abstract reasoning rather than the ability to achieve goals and wield power. Political leaders such as Trump and Stalin show high real-world authority without exceptional conceptual skills, while current AI gains narrow capabilities through targeted training.

Ken Griffin says Citadel AI agents finish PhD tasks in days

Citadel now deploys AI agents that complete complex market analysis and deal structuring tasks in days instead of months. Founder Ken Griffin shared these details in a Stanford Leadership Institute interview.

xAI completes training of its 1.5 trillion parameter Grok V9 foundation model as a major upgrade over the 0.5 trillion parameter Grok V8

Supplemental training with Cursor data comes next, followed by supervised fine-tuning and reinforcement learning ahead of a public release in three to four weeks.

Mustafa Suleyman, Microsoft AI CEO, predicts AI will reach human-level performance on most professional tasks within 18 months and fully automate white-collar roles such as accounting, legal work, marketing, and project management

A Fortune article reported the forecast and noted Suleyman's view that LLM hallucinations will largely be resolved by 2025.

Chamath Palihapitiya warns firms on direct AI model use

Consulting firms that plug OpenAI or Anthropic models straight into client work risk being displaced by those same providers. The AI companies analyze usage patterns to build competing services while gaining visibility into consultants' proprietary methods.

Richard Ngo says Effective Altruism mainly empowered Anthropic

AI researcher Richard Ngo states that Effective Altruism has mainly strengthened Anthropic by channeling funding, talent, and supportive narratives. Oliver Habryka replied that larger future impacts could come through government AI policies that do not primarily involve Anthropic.

Microsoft researcher reports Claude KV cache flush issues

A Microsoft researcher reported that Claude Code clears its KV cache after idle periods, causing performance to drop when the model resumes because it must rebuild context from tokens. This observation is consistent with results from Memento experiments on post-flush out-of-distribution behavior.

OpenAI engineer Jason Liu shares Codex workflow post

OpenAI Codex team engineer Jason Liu shared a draft guide on extending the tool to daily knowledge work such as Slidev presentations and voice note-taking. The post lists five requirements for sustained use including durable threads, memory stores, and compaction techniques with pinned megathreads.

Matthew Yglesias proposes affordable GPU quotas for data centers

A policy proposal would require new data centers to reserve 20% of GPUs for affordable compute before receiving approval. The suggestion targets greater access to scarce AI hardware during rapid infrastructure growth.

Figure's F.03 humanoid robot sorts 12,732 packages in close contest with warehouse worker

The robot completed 12,732 packages at 2.83 seconds each in a head-to-head sorting challenge with a human worker. The worker edged ahead with 12,924 packages at 2.79 seconds per package during the demonstration.

Fyodor Urnov receives flawless gene review PDF from AI tool

Gene researcher Fyodor Urnov obtained a flawless 17-page review on gene biology from an AI system within 25 minutes. Edison Scientific's deep literature mode created the PDF after a prompt on a topic the expert knew well.

Grok Imagine Generates Anime Video Of Girl Awakening Star Energy

Google AI staffer Logan Kilpatrick says model is the product

Google engineer Logan Kilpatrick stated that the AI model itself now serves as the core product. Replies from other developers reinforced the view that model capabilities deliver primary user value without added layers.

Google AI Overview Responds As Viltrumite To Invincible Search

Odyssey ML schedules live event on world models

Odyssey ML will present one or two new releases on world models in a live stream tomorrow. The 9:30 AM PT event will outline the lab's current direction and note ongoing GPU scaling work.

Gavin Leech attributes vision model size gap to data compression

Language enables much denser representations than images, which explains why current text models contain around 1000 times more parameters than vision models. This gap arises because language supports compositional semantics and high-level abstractions at greater efficiency.

Chamath Palihapitiya says Taiwan loses semiconductor importance within 18 months

Investor Chamath Palihapitiya forecasts that Taiwan's semiconductor dominance will fade in the next 18 months as the United States builds its own advanced manufacturing capacity. Critics respond that Taiwan's TSMC still produces most cutting-edge chips for AI and consumer electronics, while American plants operate at older technology levels.

Viral claims overstate Utah AI data center temperature effects

Social media posts claiming a Utah AI data center would raise nighttime temperatures by more than 20 degrees have drawn sharp criticism as physically unrealistic. Observers note that even wildfires reaching 2200°F do not produce comparable statewide warming, given data centers' far smaller heat output.

Nous Research releases Hermes Agent v0.14.0 on GitHub

Nous Research delivered a large-scale update to Hermes Agent with version 0.14.0 now on GitHub. The release covers 808 commits across 1,393 files, adds multi-agent Kanban boards and search tools, and resolves 545 issues.

Pope Leo XIV approves Vatican AI commission

Pope Leo XIV has approved the creation of a Vatican commission to study artificial intelligence's impact on people and society. The commission, coordinated by one dicastery with representatives from seven institutions, comes shortly after the pope signed an encyclical on AI set for release soon.

Meta data scientist forecasts AI abstraction advances within year

A Meta data scientist predicts AI will push tech workers toward higher abstraction layers within roughly one year. Current tools still cannot replace certain roles, though employment has long required constant adaptation to new demands.

Anthropic retires Claude 3 Opus in January 2026

Anthropic retired Claude 3 Opus in January 2026 but continues to offer access on claude.ai and through a special API. The company has approved every submitted application for that API, unlike other recent models that were fully removed.

Redis creator Salvatore Sanfilippo thanks DeepSeek AI

Redis creator Salvatore Sanfilippo publicly thanked DeepSeek AI for making advanced tools and concepts available to wider audiences. The post draws attention to upcoming V4.1 and V4.1-Vision releases expected to expand model access and capabilities.

Caitlin Kalinowski discusses AI hardware boom on Lenny's Podcast

Hardware expert Caitlin Kalinowski describes how the AI industry is shifting focus to physical systems like robots, which face supply-chain limits and potential memory price increases. She shared these insights in a Lenny's Podcast interview drawing from her roles at Apple, Meta, and OpenAI.

Greg Brockman shares Codex demo on personal messages

OpenAI's Codex analyzed three years of personal text messages and generated summaries that incorporated direct quotes, producing an emotionally resonant result for the user. Mac users can run the same process after granting access permissions to their message history.

Opus 4.7 model in Claude Code v2.1.143 fetches and filters 1350 Pokemon entries in 11 seconds

Claude's Opus 4.7 model executed a bash command to pull 1350 Pokemon entries from the PokeAPI and filter them for names ending in 'aw'. Using a 1M context window, the task completed in 11 seconds and returned the two matching results.

Max Novendstern links Thiel optimism to Ethereum and OpenAI

Max Novendstern connected Peter Thiel's idea of indefinite optimism to Vitalik Buterin's addition of a virtual machine on Ethereum and OpenAI's decision to scale transformer models. The thread frames both as major technical bets made without assured outcomes, with later replies extending the point to stochastic hardware designs.

xAI SuperGrok Integrates With Hermes Agent For Seamless Access

Nando de Freitas unveils interventional SFT for agent models

A simple change to supervised fine-tuning reduces how often language-model agents reinforce their own incorrect beliefs on factual tasks. The method trains only on observation tokens while ignoring action tokens and outperformed standard fine-tuning across dozens of prompts.

Grok Build Runs 10 Minutes By Self-Reviewing And Spawning Subagents

Dwarkesh Patel separates intelligence from goal achievement

Dwarkesh Patel argues that intelligence should be viewed as abstract reasoning rather than the ability to achieve goals and wield power. Political leaders such as Trump and Stalin show high real-world authority without exceptional conceptual skills, while current AI gains narrow capabilities through targeted training.

Ken Griffin says Citadel AI agents finish PhD tasks in days

Citadel now deploys AI agents that complete complex market analysis and deal structuring tasks in days instead of months. Founder Ken Griffin shared these details in a Stanford Leadership Institute interview.

xAI completes training of its 1.5 trillion parameter Grok V9 foundation model as a major upgrade over the 0.5 trillion parameter Grok V8

Supplemental training with Cursor data comes next, followed by supervised fine-tuning and reinforcement learning ahead of a public release in three to four weeks.

Mustafa Suleyman, Microsoft AI CEO, predicts AI will reach human-level performance on most professional tasks within 18 months and fully automate white-collar roles such as accounting, legal work, marketing, and project management

A Fortune article reported the forecast and noted Suleyman's view that LLM hallucinations will largely be resolved by 2025.

Chamath Palihapitiya warns firms on direct AI model use

Consulting firms that plug OpenAI or Anthropic models straight into client work risk being displaced by those same providers. The AI companies analyze usage patterns to build competing services while gaining visibility into consultants' proprietary methods.

Richard Ngo says Effective Altruism mainly empowered Anthropic

AI researcher Richard Ngo states that Effective Altruism has mainly strengthened Anthropic by channeling funding, talent, and supportive narratives. Oliver Habryka replied that larger future impacts could come through government AI policies that do not primarily involve Anthropic.

Microsoft researcher reports Claude KV cache flush issues

A Microsoft researcher reported that Claude Code clears its KV cache after idle periods, causing performance to drop when the model resumes because it must rebuild context from tokens. This observation is consistent with results from Memento experiments on post-flush out-of-distribution behavior.

OpenAI engineer Jason Liu shares Codex workflow post

OpenAI Codex team engineer Jason Liu shared a draft guide on extending the tool to daily knowledge work such as Slidev presentations and voice note-taking. The post lists five requirements for sustained use including durable threads, memory stores, and compaction techniques with pinned megathreads.

Matthew Yglesias proposes affordable GPU quotas for data centers

A policy proposal would require new data centers to reserve 20% of GPUs for affordable compute before receiving approval. The suggestion targets greater access to scarce AI hardware during rapid infrastructure growth.

Figure's F.03 humanoid robot sorts 12,732 packages in close contest with warehouse worker

The robot completed 12,732 packages at 2.83 seconds each in a head-to-head sorting challenge with a human worker. The worker edged ahead with 12,924 packages at 2.79 seconds per package during the demonstration.

Fyodor Urnov receives flawless gene review PDF from AI tool

Gene researcher Fyodor Urnov obtained a flawless 17-page review on gene biology from an AI system within 25 minutes. Edison Scientific's deep literature mode created the PDF after a prompt on a topic the expert knew well.

Grok Imagine Generates Anime Video Of Girl Awakening Star Energy

Google AI staffer Logan Kilpatrick says model is the product

Google engineer Logan Kilpatrick stated that the AI model itself now serves as the core product. Replies from other developers reinforced the view that model capabilities deliver primary user value without added layers.

Google AI Overview Responds As Viltrumite To Invincible Search

Odyssey ML schedules live event on world models

Odyssey ML will present one or two new releases on world models in a live stream tomorrow. The 9:30 AM PT event will outline the lab's current direction and note ongoing GPU scaling work.

Gavin Leech attributes vision model size gap to data compression

Language enables much denser representations than images, which explains why current text models contain around 1000 times more parameters than vision models. This gap arises because language supports compositional semantics and high-level abstractions at greater efficiency.

Chamath Palihapitiya says Taiwan loses semiconductor importance within 18 months

Investor Chamath Palihapitiya forecasts that Taiwan's semiconductor dominance will fade in the next 18 months as the United States builds its own advanced manufacturing capacity. Critics respond that Taiwan's TSMC still produces most cutting-edge chips for AI and consumer electronics, while American plants operate at older technology levels.

Viral claims overstate Utah AI data center temperature effects

Social media posts claiming a Utah AI data center would raise nighttime temperatures by more than 20 degrees have drawn sharp criticism as physically unrealistic. Observers note that even wildfires reaching 2200°F do not produce comparable statewide warming, given data centers' far smaller heat output.

Nous Research releases Hermes Agent v0.14.0 on GitHub

Nous Research delivered a large-scale update to Hermes Agent with version 0.14.0 now on GitHub. The release covers 808 commits across 1,393 files, adds multi-agent Kanban boards and search tools, and resolves 545 issues.

Pope Leo XIV approves Vatican AI commission

Pope Leo XIV has approved the creation of a Vatican commission to study artificial intelligence's impact on people and society. The commission, coordinated by one dicastery with representatives from seven institutions, comes shortly after the pope signed an encyclical on AI set for release soon.

Meta data scientist forecasts AI abstraction advances within year

A Meta data scientist predicts AI will push tech workers toward higher abstraction layers within roughly one year. Current tools still cannot replace certain roles, though employment has long required constant adaptation to new demands.

Anthropic retires Claude 3 Opus in January 2026

Anthropic retired Claude 3 Opus in January 2026 but continues to offer access on claude.ai and through a special API. The company has approved every submitted application for that API, unlike other recent models that were fully removed.

Redis creator Salvatore Sanfilippo thanks DeepSeek AI

Redis creator Salvatore Sanfilippo publicly thanked DeepSeek AI for making advanced tools and concepts available to wider audiences. The post draws attention to upcoming V4.1 and V4.1-Vision releases expected to expand model access and capabilities.

Caitlin Kalinowski discusses AI hardware boom on Lenny's Podcast

Hardware expert Caitlin Kalinowski describes how the AI industry is shifting focus to physical systems like robots, which face supply-chain limits and potential memory price increases. She shared these insights in a Lenny's Podcast interview drawing from her roles at Apple, Meta, and OpenAI.

Greg Brockman shares Codex demo on personal messages

OpenAI's Codex analyzed three years of personal text messages and generated summaries that incorporated direct quotes, producing an emotionally resonant result for the user. Mac users can run the same process after granting access permissions to their message history.

Opus 4.7 model in Claude Code v2.1.143 fetches and filters 1350 Pokemon entries in 11 seconds

Claude's Opus 4.7 model executed a bash command to pull 1350 Pokemon entries from the PokeAPI and filter them for names ending in 'aw'. Using a 1M context window, the task completed in 11 seconds and returned the two matching results.

Max Novendstern links Thiel optimism to Ethereum and OpenAI

Max Novendstern connected Peter Thiel's idea of indefinite optimism to Vitalik Buterin's addition of a virtual machine on Ethereum and OpenAI's decision to scale transformer models. The thread frames both as major technical bets made without assured outcomes, with later replies extending the point to stochastic hardware designs.

xAI SuperGrok Integrates With Hermes Agent For Seamless Access

Nando de Freitas unveils interventional SFT for agent models

A simple change to supervised fine-tuning reduces how often language-model agents reinforce their own incorrect beliefs on factual tasks. The method trains only on observation tokens while ignoring action tokens and outperformed standard fine-tuning across dozens of prompts.

Grok Build Runs 10 Minutes By Self-Reviewing And Spawning Subagents