AI Weekly - Week 51/2025
TL;DR
This week in 30 seconds:
- Reality Check: MIT Technology Review declares 2025 the year of “AI Hype Correction”
- Job Crisis: 55,000 layoffs in the US were explicitly attributed to AI
- Regulation: NY signs RAISE Act – first state law requiring AI transparency for frontier models
The Story of the Week
2025: The Year of AI Reckoning
MIT Technology Review delivered a damning verdict this week: 2025 was “a year of reckoning” for AI. [1] The central thesis: LLMs are not the path to AGI.
This is no fringe opinion. The arguments:
-
Scaling Laws Plateau: More compute doesn’t deliver proportionally better results anymore. The low-hanging fruit has been picked. [1]
-
Practical Limitations Become Visible: Hallucinations persist. Reasoning remains superficial. And the METR study shows: Experienced developers work with AI tools 19% slower, not faster. [8]
-
Business Value Fails to Materialize: Investments are massive – OpenAI alone has $1.4 trillion in infrastructure commitments. But the returns? MIT cites studies showing: 95% of enterprise AI pilots deliver no measurable value. [1]
“The tricky thing about AI is that it is amazing and it is also highly limited.” - Andrew Ng, Turing Award Winner [6]
Ng’s assessment is particularly striking because he’s one of the biggest AI enthusiasts. When even he says “highly limited,” you should listen. [6]
What This Means: Not that AI is useless – but that expectations have outpaced reality. The hype cycle is turning toward the “Trough of Disillusionment.” If you’re investing in AI now, you should have concrete use cases, not “AI for everything.” [1]
Bottom Line: 2025 was the year AI transitioned from “magic” to “tool.” That’s not bad – it’s maturation. But for those expecting AGI in 2-3 years, it’s a cold shower.
More Top Stories
55,000 Jobs: AI as Reason for Termination
The numbers are in: 55,000 layoffs in the US were explicitly attributed to AI in 2025. [2] According to Challenger, Gray & Christmas, this is the highest count since tracking began. [2]
The Biggest Cuts: [2]
- Amazon: 14,000 corporate roles (October) – largest workforce reduction in company history
- Microsoft: ~15,000 jobs throughout 2025, including 9,000 in July alone
- Salesforce: 4,000 customer support roles replaced by Agentforce
CEO Marc Benioff was blunt about it: “AI has flattened hiring needs.” [2] Salesforce’s Agentforce handles customer support that humans previously managed.
But Context Matters: Andy Jassy (Amazon CEO) emphasized the cuts were “not really AI driven – not right now, at least.” [2] The truth probably lies somewhere in between: AI is a welcome justification for restructurings that were planned anyway.
The Irony: While AI eliminates jobs, it also creates new ones. But the new jobs require skills that laid-off workers often don’t have.
NY RAISE Act: Transparency Becomes Mandatory
On December 19th, Governor Kathy Hochul signed the RAISE Act – the first US state law mandating transparency for frontier AI models. [3]
What Developers Must Now Do: [3]
- Publish safety protocols
- Report incidents to the state within 72 hours
- Accept oversight from a new office within the Department of Financial Services
The law targets “large AI developers” – meaning OpenAI, Anthropic, Google, Meta. [3] Smaller startups are (for now) exempt.
The Tension: Just 8 days earlier, Trump signed an executive order against “onerous” state AI laws. NY’s RAISE Act will likely be one of the first tests of this federal vs. state confrontation.
Anthropic: DOE Partnership for Science
Anthropic and the US Department of Energy announced the “Genesis Mission” – a multi-year partnership to bring AI to all 17 national laboratories. [4]
Three Focus Areas: [4]
- Energy Dominance: Optimizing grids, renewables, nuclear
- Life Sciences: Drug discovery, protein folding, genomics
- Scientific Productivity: Automating repetitive research tasks
This is strategically clever: While competitors fight over the consumer market, Anthropic positions itself as “AI for Science.” Government contracts provide long-term stability and are less hype-dependent.
In Parallel: Anthropic published its compliance framework for California’s SB53 – proactively, before the law takes effect. [5] This signals: We take regulation seriously.
Quick Hits
Quick notes:
- Gemini 3 Flash Default: Google makes Flash the standard in Gemini App + Search globally. 3x faster than 2.5 Pro. Launch
- ChatGPT Images 1.5: 4x faster, improved instruction following, consistent facial likeness. OpenAI
- Meta Smart Glasses “Hear Better”: AI noise suppression isolates speakers in loud environments. News
- OpenAI Academy for News Orgs: Tools, grants, support for journalists. OpenAI
- Anthropic Skills Open Standard: Repeatable workflows now usable across platforms. SD Times
- Biotech AI Milestone: Multiple AI-discovered drugs reach Phase 2/3 trials – 2026 will be a landmark year. News
Tool of the Week
ChatGPT Images 1.5 - Image generation reimagined
OpenAI has completely overhauled its image generator. The key improvements:
- 4x faster than the previous version
- Instruction following dramatically improved – complex prompts now work reliably
- Facial likeness remains consistent through edits
- Lighting is properly preserved
The use case: product mockups, social media content, presentations – anything where you need visual assets quickly, without a designer.
Fail of the Week
Stanford Study: 20% fewer junior developer jobs since 2022
While the industry talks about “AI augmentation,” the numbers tell a different story: Employment of software developers aged 22-25 has declined by 20% since 2022. [7]
Correlation isn’t causation – but the timing is suspicious. GitHub Copilot launched in 2022. ChatGPT launched in late 2022. Claude Code launched in 2025. [7]
The Problem: Entry-level jobs are often the ones AI can replace best – repetitive code, bug fixes, documentation. But these exact jobs are the gateway to a career. [7]
Root Cause: The industry is optimizing for short-term efficiency instead of long-term talent pipeline.
What We Learn: If we eliminate junior positions, where do the seniors come from in 5 years?
Number of the Week
55,000
AI-driven layoffs in the US in 2025 according to Challenger, Gray & Christmas [2]
This isn’t an “AI replaces all jobs” scenario – it’s 55,000 out of over 160 million. But it’s a trend. And for those affected, it’s 100% of their reality. The question isn’t whether AI changes jobs, but how fast and who supports the transition.
Reading List
For the weekend:
- The great AI hype correction of 2025 - MIT TR’s comprehensive analysis of why 2025 was a turning point (12 min)
- Andrew Ng: AI is limited - Turing Award winner on realistic AI expectations (6 min)
- AI job cuts 2025 - CNBC’s breakdown of layoff figures (8 min)
Next Week
What’s coming:
- Christmas break – but AI never rests
- Nvidia-Groq Deal finalization expected
- CES 2025 Preparations – hardware announcements starting January 7th
- Year-in-Review Content from all major players
This newsletter was AI-assisted in research and writing. Hero image generated with Pollinations.ai.