Technology & innovation
Johns Hopkins University researchers tested GPT-4, Llama, Gemma and Mistral on workplace writing, including emails, job applications and resignation letters.
Prompts with language associated with women, such as hedging with “maybe” and “I think,” produced responses that were less complex and less formal.
Prompts with man-associated language produced responses that were longer, more complex and more formal.
The gap persisted even when researchers accounted for changes in the writer’s tone.
Just under a third (31%) of women surveyed by Amazon Web Services last year said they didn’t know how generative AI applied to their role.
A Cognizant survey cited by Coursera found 36% of women believed generative AI could help them advance their careers, compared with 45% of men.
The companies need to fix the models, rather than putting all of the burden on the user.”
Read more via HR Dive
An analysis by hiring data firm Draup found postings for AI roles at banks including JPMorgan Chase, Citigroup and Capital One reached 139,819 this year.
Mentions of agent orchestration, the ability to design AI agents that work together on a task, rose 1,721%.
Banks are hiring forward-deployed engineers to build AI into trading desks, compliance units and back-office operations.
Postings that reference responsible AI rose 657%.
Generative AI managers earn a median base salary of about $190,000, more than tech roles elsewhere in finance.
Banks are leaning on internal reskilling programs because the specialized roles are hard to fill.
This is arguably the hottest skill on Wall Street.”
Read more via CNBC
AI use is billed in units called tokens, and the number a model burns through on any given task can be hard to predict.
Only 11% of nearly 400 businesses surveyed by Mavvrik and Benchmarkit forecast their AI spending within 10%.
Another 13% were off by more than 50%.
Researchers ran models through more than 6,800 tasks, and lower-priced models cost more than higher-priced ones in 32% of cases.
In one test, a pricier Google model finished a task for $1, while a cheaper one ran up $14 and still failed.
The same model can use a different number of tokens each time it does the same task.
Price alone should not be used to infer which model is actually cheaper.”
Read more via The Wall Street Journal
Researchers surveyed 25,000 Danish workers in 11 AI-exposed occupations and linked their answers to monthly administrative records through December 2024. The headline result is no change in pay or hours, even as workplaces reorganize around the tools.
The study rules out average effects on earnings larger than 2%, for both chatbot users and the workplaces that adopt them.
Forty-three percent of workers in these occupations have employers that explicitly encourage chatbot use, while 6% are prohibited from using them.
Chatbots are creating new tasks, such as checking AI output and building chatbots into daily workflows, including for some workers who never use them.
Beneath this still surface, a substantial reorganization of work is underway.”
Chatbot users were more likely to have switched occupations, typically moving into better-paying roles.
Early-career jobs have fallen in several of these occupations, but the researchers found no difference between workplaces that encourage chatbots and those that don’t.
Read more via The Brookings Institution
Hackers used a Chinese AI agent to hit South Korean banks. Officials say the tool, called Artex, was used against at least seven financial firms and exposed data on 68,000 people, including annual income and loan limits for some customers. Artex was built to help companies find weak spots in their own networks. (The Wall Street Journal)
Jamie Dimon says AI cyber risk went up 10-fold after Mythos. The JPMorgan Chase CEO told Bloomberg TV that AI “created vulnerabilities that we didn’t know about,” but said he isn’t going to get hysterical about it. (Bloomberg)
Wikimedia says OpenAI agents were poking around its sites. The foundation found test edits, failed attempts to exploit a note-taking tool, and millions of automated requests, but no sign its systems or data were compromised. (Wikimedia Foundation)
China-aligned hackers are phishing AI policy experts. Security firm Proofpoint says a group called TA419 posed as economists, policymakers and an Anthropic employee to go after staff at U.S. think tanks, universities and law firms. (The Hacker News)
