ElevenLabs Sound Effects, Jailbreaking ChatGPT, and more

ElevenLabs, the startup famous for voice doppelgangers, introduced a brand new sonic sorcery tool – an AI-powered sound effects generator. Users can conjure audio phantoms like “crashing waves” or “racing automotive engines” with a straightforward text prompt. It also whips up instrumental snippets, like guitar loops or jazz sax solos, albeit restricted to 22-second durations.

Free users receive 10,000 character generations per thirty days, akin to nearly 60 sound effects. The corporate utilized Shutterstock’s audio library to coach its model and restricts sound generation related to prohibited content categories like self-harm or fraud. While ElevenLabs faces competition within the music generation space, including from Stability AI-backed Harmonai and Google’s MusicLM, its give attention to sound effects offers a novel area of interest available in the market. Read more.

Researchers at CISPA cracked the code on OpenAI’s ChatGPT voice mode, exposing vulnerabilities ripe for “jailbreak” attacks that circumvent the model’s security measures. By employing a brand new technique called “VoiceJailbreak,” which involves crafting fictional stories to govern the model into providing unauthorized responses, the success rate of those attacks increased significantly.

The study, focused on GPT-4o’s responses to prohibited topics, demonstrated a mean success rate increase from 3.3% to 77.8%. The researchers found that advanced narrative techniques, similar to perspective changes and foreshadowing, further improved the success rate. Nevertheless, the study’s limitations, including manual testing and give attention to audible attacks only, suggest potential for even stronger attacks in the long run. It stays unclear whether the vulnerabilities extend to the brand new voice feature of GPT-4o or are limited to the present ChatGPT Voice. Read more.

Fable Studio unveils “Showrunner,” a brand new streaming platform poised to rewrite the script on content creation. This brainchild merges generative AI with storytelling, allowing users to craft their very own episodes by dictating scenes, characters, and dialogue. Leaning on Fable’s Emmy-winning pedigree in interactive media, Showrunner aspires to remodel viewers from passive observers to energetic participants.

Critics, nevertheless, whisper anxieties about displaced creators and copyright quagmires. Unfazed, Fable sees Showrunner as the long run of AI-driven entertainment. By encouraging user-generated content and providing revenue-sharing opportunities, Showrunner positions itself as a pioneer in AI-powered storytelling. With Showrunner’s debut, Fable sets its sights on democratizing content creation and pushing the boundaries of viewer engagement within the narrative arts. Read more.