- Weekly GenAI
- Posts
- MIT Says ChatGPT Could Be Making You Dumber
MIT Says ChatGPT Could Be Making You Dumber
Let's get started.
MIT Study Warns: ChatGPT May Undermine Critical Thinking
A new MIT Media Lab study suggests that using ChatGPT for academic tasks may impair brain activity and critical thinking, especially among young users. In an experiment with 54 participants writing SAT-style essays using ChatGPT, Google Search, or no tools, those relying on ChatGPT showed the lowest cognitive engagement. EEG scans revealed diminished neural activity, and essays were rated as formulaic and lacking originality.
Over time, ChatGPT users relied more heavily on copy-paste tactics and retained less information. In contrast, those writing unaided or with Google demonstrated higher brain connectivity and satisfaction. Researchers warn that overdependence on AI may hinder long-term learning and memory development.
While the study is still awaiting peer review and has a small sample size, its authors urge caution and call for stronger oversight before integrating generative AI into education.
Introducing Midjourney’s V1 Video Model
Midjourney is taking a major step forward: Version 1 of our Video Model is now live. This marks the transition from static imagery to moving visuals, paving the way toward our long-term vision—real-time, open-world AI simulations.
What’s new?
The “Image-to-Video” workflow lets users animate images directly within Midjourney. You can choose from:
Automatic animation, which generates motion on its own
Manual mode, where you guide how the scene should move
Low motion, ideal for subtle scenes
High motion, best for dynamic, action-packed visuals
You can also animate external images by uploading and applying a motion prompt.
Each job generates four 5-second videos, and you can extend them up to 20 seconds total. The cost is about 8x an image job, or roughly one image-worth per second of video—25x cheaper than current market options.
For now, this feature is web-only, with video relax mode available to Pro subscribers and above.
Meta Tried to Buy Ilya Sutskever’s AI Startup—Now It's Hiring His Team
Meta recently attempted to acquire Safe Superintelligence, the $32B AI startup founded by Ilya Sutskever, but was rejected. Instead, Mark Zuckerberg is hiring CEO Daniel Gross and Nat Friedman, co-founders of the startup and the VC firm NFDG, in which Meta will take a stake.
The hires follow Meta’s $14.3B investment in Scale AI and reflect its aggressive push into AGI. Gross and Friedman will join Meta’s AI team under Alexandr Wang.
OpenAI’s Sam Altman says Meta has offered $100M+ signing bonuses to lure talent, but claims top employees have stayed. Meta says more details are coming on its superintelligence effort.
Inside Nvidia’s Expanding AI Empire
No company has ridden the AI wave harder than Nvidia. Since the debut of ChatGPT, its revenue, stock, and cash reserves have soared — and so has its appetite for AI startup investments.
In 2024, Nvidia joined 49 funding rounds for AI companies, up from 34 in 2023 — more than its prior four years combined. Its corporate VC arm NVentures has also ramped up, participating in 24 deals in 2024, up from just 2 in 2022.
Notable investments include:
OpenAI – $100M in a $6.6B round
xAI – Participated in Elon Musk’s $6B round
Inflection AI – $1.3B (later partially acquired by Microsoft)
Scale AI – $1B with Meta and Amazon
Wayve – $1.05B for autonomous driving
Figure, Mistral, Cohere, Perplexity, Lambda, Poolside – Rounds ranging from $480M to $675M
CoreWeave, Together AI, Waabi, Crusoe, Imbue – Infrastructure and cloud-focused AI investments
Hippocratic AI, Sandbox AQ, Weka, Runway, Bright Machines – Sector-specific AI startups (healthcare, quantum, media, robotics)
Nvidia’s strategy: back “game changers and market makers” to expand the AI ecosystem—often supporting direct rivals to OpenAI, or companies integrating Nvidia GPUs into their services.
With over 30 large AI investments since 2023, Nvidia is not just powering AI — it’s shaping the next generation of it.
AMD Launches MI350 AI Chips to Challenge Nvidia, Unveils Developer Cloud
AMD has introduced its new MI350X and MI355X AI chips, designed to rival Nvidia’s Blackwell processors. The company claims up to 4x AI performance and a 35x increase in inferencing capability over its previous generation.
Each chip features 288GB of HBM3E memory, with configurations scaling up to 128 GPUs and 2.3TB of memory, using either air or liquid cooling.
AMD also previewed its upcoming MI400 series, set to launch in 2026, featuring up to 432GB of HBM4 memory and speeds of 19.6TB/s.
In addition, AMD announced AMD Developer Cloud, a new service providing remote access to its MI300 and MI350 GPUs for AI training and inference — a direct competitor to Nvidia’s DGX Cloud.
Despite the innovation, AMD’s stock is down 24% over the past year, while Nvidia is up 19%. Both companies have been hit by U.S. export restrictions to China: AMD expects an $800 million loss, while Nvidia reported $4.5 billion in write-downs and $8 billion in missed sales.