🗞 OpenAI's Training Controversy, Meta's Leadership Change, and Anthropic's AI Microscope

🔎 The Latest on the AI Frontier:

• OpenAI's GPT-4o Suspected of Training on Paywalled Books

• Meta's AI Research Chief Exits During Major Investment Push

• Anthropic Creates "AI Microscope" to Study Language Model Reasoning

• ChatGPT Shows Human-Like Decision Biases in New Study

• White House Meeting to Decide TikTok's Fate as Ban Deadline Looms

• Other news you might find interesting

📊 Study claims OpenAI may have trained GPT-4o on paywalled O'Reilly book content

• Research from AI Disclosures Project suggests GPT-4o shows "strong recognition" of paywalled O'Reilly Media books compared to older models like GPT-3.5 Turbo, despite no licensing agreement with O'Reilly.

• Using a methodology called DE-COP, researchers analyzed 13,962 paragraph excerpts from 34 O'Reilly books to estimate the probability of content inclusion in training datasets.

• The study isn't conclusive - researchers acknowledge limitations in their experimental method and didn't evaluate OpenAI's newest models including GPT-4.5, o3-mini, and o1.

👩‍💻 Meta AI research leader Joelle Pineau announces departure amid company's major AI investment push

• Pineau, who has headed Meta's Fundamental AI Research (FAIR) since 2017, announced her resignation as the company continues to prioritize AI development.

• Meta CEO Mark Zuckerberg has designated AI as the company's top priority, planning to spend up to $65 billion on AI efforts this year as part of a broader industry trend.

• Meta, Microsoft, and Amazon are projected to collectively spend $371 billion on AI data centers and computing resources in 2025—a 44% increase from last year—and $525 billion annually by 2032.

🔬 Anthropic researchers use "AI microscope" to understand how language models think

• New research from Anthropic is shedding light on LLM reasoning, with scientists developing tools that track data patterns and information flows within AI systems to observe how they connect concepts.

• The study reveals AI systems approach problems differently than humans, using probabilistic reasoning rather than step-by-step calculation—for example, when adding 36 and 59, the AI approximates the answer and estimates the last digit simultaneously.

• Researchers are also investigating whether multilingual models "think" in the language of user prompts or use universal concepts regardless of language, addressing a key question in understanding AI's internal processes.

🧠 Study finds ChatGPT exhibits human-like decision biases in nearly half of tests conducted

• New research published in Manufacturing & Service Operations Management reveals ChatGPT displays human cognitive biases like overconfidence, ambiguity aversion, and conjunction fallacy in judgement-based tasks.

• The AI excels at logical and probability problems but struggles with subjective reasoning, with findings showing biases remained stable across business situations but varied between model versions.

• Researchers recommend businesses treat AI "like an employee who makes important decisions" by implementing regular audits and ethical guidelines to prevent automating flawed thinking.

🔄 TikTok's future to be discussed in key White House meeting as ban deadline approaches

• Trump administration officials including VP JD Vance, Commerce Secretary Howard Lutnick, and others will meet Wednesday to review TikTok's sale options to avoid the April 5 ban deadline.

• Private equity firm Blackstone and venture capital firm Andreessen Horowitz are reportedly considering minority investments alongside existing non-Chinese shareholders to help ByteDance meet US requirements.

• Pew Research survey shows declining concerns about TikTok, with 49% of US adults viewing it as a national security threat (down from 59% in 2023) and only 34% supporting a ban (down from 50%).

More news you might find interesting:

• Meta becomes UFC's Official Fan Technology Partner in multiyear deal.

• Nvidia open sources KAI Scheduler to improve AI workload management.

• Harvard study reveals individuals using AI match performance of two-person teams without it.

• Microsoft closes Shanghai AI lab, signaling broader retreat from Chinese market.

• Microsoft enhances Copilot with experimental features currently being tested.

• Tech experts warn increasing AI dependence could erode core human skills.

• Father uses AI to accelerate research for son's rare, fatal disease.

• Spain's LaLiga embraces AI as fundamental to its business strategy and fan engagement.

• Chinese brain chip company on track to surpass Neuralink in human implants this year.

• Samsung reportedly developing solid-state battery for Galaxy Ring 2 to improve battery life.

• Nintendo expected to reveal Switch 2 details at Nintendo Direct event today.

Have any feedback? Send us an email