Meredith Broussard: just a software update

Meredith Broussard said OpenAI’s GPT-5.5 is “just a software update,” pushing back on the language surrounding the company’s April 23 launch. (youtube.com) (openai.com) Broussard is a data journalism professor at New York University and research director at the NYU Alliance for Public Interest Technology. She is also the author of *More Than a Glitch* and *Artificial Unintelligence*. (meredithbroussard.com) OpenAI introduced GPT-5.5 as “a new class of intelligence for real work” and said the model is built for coding, online research, spreadsheets, and software use. The company said GPT-5.5 began rolling out in ChatGPT and Codex on April 23, with an API release added April 24. (openai.com 1) (openai.com 2) The company’s pitch leaned heavily on “agentic” work: giving the model a messy task and letting it plan, use tools, and keep going. Broussard said that framing can make a probabilistic text system sound more dependable than it is. (openai.com) (youtube.com) A large language model predicts likely next words from patterns in training data. That can produce fluent answers, but it can also produce confident mistakes, which the industry calls hallucinations. (artificialanalysis.ai) (undark.org) OpenAI’s release post highlighted benchmark gains, including 82.7% on Terminal-Bench 2.0 and 78.7% on OSWorld-Verified. Broussard said benchmark selections can be narrow enough to flatter a model and leave users with a distorted picture of reliability. (openai.com) (youtube.com) Independent testing points to that gap. Artificial Analysis said GPT-5.5 scored 57% accuracy on its AA-Omniscience benchmark and a score of 20 on the benchmark’s combined factuality index, which penalizes bad guesses and rewards saying “I don’t know.” (artificialanalysis.ai) OpenAI said it ran GPT-5.5 through its full predeployment safety evaluations, targeted red-teaming, and feedback from nearly 200 early-access partners. The company said it released the model with its “strongest set of safeguards to date.” (openai.com) Broussard has been making a related argument for years under the label “technochauvinism,” her term for assuming the newest technical system is automatically the best answer. In her 2018 book, she argued that people had “stopped demanding that our technology actually work.” (publicbooks.org) (mitpress.mit.edu) In a 2023 interview with *Undark*, she said AI debates should focus less on imagined futures and more on harms already visible in deployed systems, including race, gender, and disability bias. That same critique sits underneath her reaction to GPT-5.5. (undark.org) Her point was not that GPT-5.5 changed nothing. It was that a faster, more capable model is still software, and software still has to be tested, audited, and distrusted when the stakes are high. (youtube.com) (openai.com)