‘A.I.-Washing’ Layoffs? + Why L.L.M.s Can’t Write Well + Tokenmaxxing

Episode Summary

Executive Summary: This episode of Hard Fork covers three main topics: the rise of AI-related tech layoffs at companies like Atlassian, Block, and Meta; the limitations of AI in creative writing, with guest Jasmine Sun explaining why LLMs struggle with literary quality due to post-training and lack of lived experience; and the emergence of 'token maxing' leaderboards in tech companies that track employees' AI token usage, raising concerns about productivity metrics and incentives.

Main Topics: AI-Related Tech Layoffs (Priority: 5/5): Discussion of recent layoffs at Atlassian (10% reduction), Block (40% reduction), and potential Meta layoffs (up to 20%), with CEOs citing AI as a factor. Analysis of whether these are genuine AI-driven changes or 'AI washing' to justify cuts. AI and Creative Writing Limitations (Priority: 4/5): Guest Jasmine Sun argues that while LLMs excel at text generation, they struggle with literary writing due to post-training that enforces blandness, lack of verifiable rewards, and absence of lived experience. She notes that earlier models like GPT-2/3 were more creative. Token Maxing and Leaderboards (Priority: 4/5): Tech companies are creating leaderboards to track employees' AI token usage, with top users consuming billions of tokens. This creates perverse incentives and raises questions about productivity measurement and cost. AI Washing vs. Genuine AI Impact (Priority: 3/5): Debate over whether companies are using AI as an excuse for layoffs or if AI truly changes workforce needs. Examples include Block's Jay-Z event spending and Meta's AI infrastructure investments. Human-AI Collaboration in Writing (Priority: 3/5): Jasmine Sun describes using Claude as an editor by feeding it her writing archive to create personalized rubrics, finding it helpful for feedback without replacing her voice.

Key Arguments: AI is being cited as a reason for layoffs, but some cases may be 'AI washing' to justify cuts and boost stock prices. LLMs are good at text generation but poor at literary writing due to post-training that prioritizes helpfulness over creativity. Token usage leaderboards incentivize wasteful consumption and may not correlate with actual productivity. Earlier AI models (GPT-2/3) were more creative because they lacked post-training constraints. Human-AI collaboration can enhance writing if the AI is tuned to the writer's personal style and goals.

Data Points: Atlassian layoff percentage: 10% - About 1,600 jobs cut to fund AI investment. Block layoff percentage: 40% - About 4,000 jobs cut, citing shift to smaller teams. Potential Meta layoff percentage: 20% - Up to 16,000 jobs, reported by Reuters. Block event spending: $68 million - Spent to fly 8,000 people to an in-person event with Jay-Z. Meta AI infrastructure spending: $135 billion - Capital expenditures planned for this year. Top token user at OpenAI: 210 billion tokens - Over a seven-day period, equivalent to about 33 Wikipedia's worth of text. Top Claude Code user spending: $150,000 - Spent on tokens in one month.

Pivotal Quotes: "We're not making this decision because we're in trouble. Our business is strong, but something has changed. I had two options: cut gradually over months or years as this shift plays out, or be honest about where we are and act on it now." — Jack Dorsey: Explaining Block's layoffs, citing AI as a factor. "Projects that used to require big teams now can be accomplished by a single very talented person." — Mark Zuckerberg: On Meta's earnings call, justifying potential layoffs. "The thing that really shocked me is that, like, in a way, the writing style of GPD2 and GPD-3, I found so much more compelling than ChatGPT today." — Jasmine Sun: Discussing how earlier AI models were more creative in writing.

Implications: AI is reshaping tech employment, with layoffs and token tracking becoming common. Creative writing may remain a human stronghold, but AI collaboration tools are evolving. Token leaderboards risk creating perverse incentives, but they signal a shift toward AI-centric productivity metrics.