ChatGPT’s ‘Study Together’ Aims to Reinvent Learning
With CoreWeave Scaling and LLMs Learning to Remember
TLDR; In today’s Signal
CoreWeave to Acquire Core Scientific in $9B
OpenAI’s Stock-Based Compensation Soared to $4.4B
ChatGPT’s ‘Study Together’ Mode Emerges
Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions
OpenAI ramps up security amid threats
AI is accelerating data-world consolidation
Eudia Expands With First Acquisition After $105M Series A
Robotera Raises $70M Series A
THE AI SIGNAL PICKS
CoreWeave to Acquire Core Scientific in $9B to Expand AI Data Center Capacity
CoreWeave has announced a ~$9 billion all-stock acquisition of bitcoin mining firm Core Scientific, aiming to expand its AI data center footprint by adding 1.3 GW of power capacity. The deal, expected to close in Q4 2025, marks a strategic shift from crypto to AI infrastructure and is projected to save CoreWeave over $10 billion in future lease costs.
OpenAI’s Stock-Based Compensation Soared to $4.4B Amid Intense AI Talent Wars
In 2024, OpenAI’s stock compensation surged five-fold to $4.4 billion, equating to 119% of its total revenue, a vivid indicator of the steep talent competition in AI. Facing aggressive recruitment from Meta, which reportedly offered $100 million bonuses to several researchers, OpenAI has expanded equity ceilings, stock refresh grants, and boosted cash bands, some principal researchers now earn up to $975 000 in base salary
Cursor Scrambles After User Outcry Over Hidden, Usage-Based Pricing Model
Cursor, the AI-powered code assistant, faced a major backlash for quietly shifting its “unlimited” Pro plan to a usage-based model tied to API costs. Users noticed drastically faster depletion of tokens, some depleting a month's worth in days, prompting a wave of subscription cancellations and criticism of poor communication. In response, Cursor issued a public apology, refunded unexpected charges, clarified their $20 monthly usage credit for frontier models with grace periods, and promised better transparency and dashboard visibility.
Anthropic Pushes for AI Lab Transparency
Anthropic has introduced a Transparency Framework targeting frontier AI development, urging labs to share risk assessment strategies, system documentation, and employee protections to strengthen accountability and safety protocols.
THE BIG LEAP
ChatGPT’s ‘Study Together’ Mode Emerges in Tests as Interactive Learning Companion
Signal Scoop: OpenAI is quietly rolling out a new experimental feature in ChatGPT called “Study Together”, a tool that reimagines how AI can support real learning. Instead of just answering questions, this mode actively engages users in step-by-step learning, simulating the experience of studying with a helpful peer or tutor.
The Full Picture
“Study Together” shifts ChatGPT into a more interactive and co-learning mode. Rather than passively giving answers, the AI prompts users with quizzes, follow-up questions, challenges, or guided explanations, tailored to their learning goals.
The feature aims to help users think critically, retain knowledge, and avoid passive consumption. It seems to encourage actual understanding—especially useful for students or lifelong learners.
Users in A/B tests have noticed the feature in the ChatGPT tools menu, particularly in Plus accounts, though it appears inconsistently. Some speculate it could eventually include multi-user collaboration, similar to group study sessions.
OpenAI is likely responding to criticism that ChatGPT enables homework cheating and shallow learning. This mode could help reframe ChatGPT as an ethical, active learning partner, much like Google's recent LearnLM initiative.
OpenAI hasn’t officially announced “Study Together.” It’s quiet rollout suggests it’s still experimental, with user feedback likely shaping its evolution.
What You Can’t Miss
This is a major shift from answer-based AI to education-focused engagement. ChatGPT isn’t just telling you the answer, it’s helping you figure it out. If successful, this could disrupt traditional tutoring apps and even reshape digital classroom tools. Only select ChatGPT users can see “Study Together,” and OpenAI hasn’t shared when (or if) it will launch broadly.
Evaluating Memory in LLM Agents via Incremental Multi-Turn Interactions
Signal Scoop: As large language models (LLMs) evolve into long-term agents, their ability to remember, update, and reason over time becomes critical. This paper introduces a benchmark—MemoryAgentBench—to systematically evaluate these memory capabilities in incremental multi-turn settings.
The Full Picture
Designed to test memory in realistic, multi-turn agent scenarios.
Focuses on incremental memory load, agents are fed info step-by-step and must retrieve, update, or reason over it.
Recall specific facts from earlier turns, learn and apply new knowledge introduced during interaction, summarize or reason over long contexts, detect and resolve contradictions or updates to previous memory are the core memory skills
Models and systems evaluated are long-context models like Claude, GPT-4, etc., with large token windows, Retrieval-Augmented Generation models, MemGPT, Mem0, and other open-source memory agents.
What You Can’t Miss
Simply increasing token limits doesn’t replicate true memory. Real memory includes updating, forgetting, and resolving contradictions. By breaking memory into AR, TTL, LRU, and CR, it gives the field concrete dimensions to improve on. This kind of memory, incremental, interactive, self-correcting, is exactly what future agent-based LLMs will need for tasks like personal assistance, tutoring, or long-term planning.
ON THE AI EDGE
OpenAI ramps up security amid threats
OpenAI has ramped up security by suspending numerous ChatGPT accounts linked to state-affiliated threat which were found exploiting the platform for activities like malware development (e.g., the “ScopeCreep” trojan), social media manipulation, disinformation campaigns, and employment scams
AI is accelerating data-world consolidation
As deals like Databricks acquiring Neon ($1 B) and Salesforce buying Informatica ($8 B) show, companies are snapping up best-in-class data tech to build robust data foundations essential for AI deployments. For years, the data landscape has been overpopulated with point solutions, ETL, reverse ETL, metadata tools, etc, making it hard to integrate data coherently.
AI Is Shaking Up Hollywood: Startups & Investors Dive In
AI is radically transforming Hollywood—from special effects to dubbing, distribution, and even financing. A wave of startups (like Runway, Ecco, Chronicle Studios, Toonstar, Deepdub, Moments Lab) are deploying AI to streamline animation, editing, localization, marketing, and content discovery
AI START-UP NEWS
Eudia Expands With First Acquisition After $105M Series A
Eudia emerged from stealth in February 2025 with a Series A of $105 million, led by General Catalyst, $30 million upfront and $75 million contingent on acquisitions.
Robotera Raises $70M Series A to Accelerate Humanoid Robotics R&D & Production
Robotics startup Robotera, spun out of Tsinghua University, has secured CNY 500 million (~US $70 million) in Series A funding to further scale its advanced embodied AI robotics business.
Synfini Scores $8.9M to Supercharge AI-Powered Drug Discovery
Synfini, an emerging AI-driven drug discovery startup, raised $8.9 million in its latest round led by JSL Health Capital, bringing its total funding to $53 million through a mix of venture capital and non-dilutive grants.
NEW TOOLS, NEW POSSIBILITIES
Magnific AI: A tool for upscaling images.
Humata: Upload documents and then ask it questions
Intellectia.AI: A tool for investment research and analysis.
RambleFix: A tool to convert spoken thoughts into written text.
Kaiber: Video generation engine, using your own images or text
Opus Clip: A tool to create short clips.
Windsurf: Helps developers understand, modify, and test unfamiliar codebases using natural language
AI CAREER HORIZON
micro1: Computer Scientist (Remote, U.S)
Jobright.ai: Software Engineer (Remote, U.S)
Joinrs US: Junior Web Developer (Remote, U.S)
G2i Inc.: Software Engineer, AI (Remote, U.S)
Elevate your experience. Join our community
Please help us get better and suggest new ideas at ceo@theaisignal.com