Voice mimicking in GPT-4o, simplified access to Midjourney, and AI products battling for public affection — the top 3 AI news stories of the week
Our latest AI Digest covers the biggest breaking AI news of the week. Anywhere Club community leader, Aliaksei Kartynnik, comments on key stories.
#1 — Slow down, OpenAI is recording
OpenAI has released a report on the testing of GPT-4o for vulnerabilities and overall security. The report identifies unauthorized voice generation as a key area of risk evaluation and mitigation. It is concerning that GPT-4o can mimic user voices, including “rare” instances in which it does so unprompted. Imagine the potential benefits to malefactors if they could steal such a model or its data? Should we still trust OpenAI?
#2 — Midjourney is more accessible
Access to image generation in Midjourney no longer requires registration on Discord. The company’s website now features image generation functionality directly in the browser, and it is possible to log in not only through Discord, but also via a Google account. You can make up to 25 images for free, and there is a new subscription available for $10 per month that includes up to 200 image generations each month.
#3 — Ranking: which AI products are in demand?
Venture capital firm Andreessen Horowitz has published its third “Top 100 Gen AI Consumer Apps” ranking. It can be useful to know which companies and ideas are popular. The top five leaders in web products, based on unique monthly visits to the sites, are (perhaps as expected): ChatGPT, Character.ai, Perplexity, Claude, and SUNO. The top three leaders in mobile apps, based on active users per month, are more surprising: ChatGPT, Microsoft Edge, and Photomath. Interesting to see that a forgotten browser and an app for solving handwritten mathematical equations ranked so highly. Wow.
.png)
