41.7K subscribers
5.53K photos
232 videos
5 files
917 links
🤖 Welcome to the ChatGPT telegram channel! Here, we post the latest news, updates, and examples of using the ChatGPT large language model for generating human-like text in conversations. Subscribe to stay up-to-date and learn more about its capabilities.
Download Telegram
More of “a” x 1000

It’s like when you’re in a dream, but suddenly it swerves in some random direction.

Looking into ChatGPT’s uncontrolled dreams.
🤯8🤣63
Concatenate '<|endo with 'ftext|>'
🤣12👏2
FYI: Lifted bans on the ~300 that had been auto-blocked by Rose in the group.
👍43
Facebook Open Sources AudioCraft: Generative AI for Audio

“Today, we’re excited to release an improved version of our EnCodec decoder, which allows for higher quality music generation with fewer artifacts; our pre-trained AudioGen model, which lets you generate environmental sounds and sound effects like a dog barking, cars honking, or footsteps on a wooden floor; and all of the AudioCraft model weights and code. The models are available for research purposes and to further people’s understanding of the technology.”

Announcement

Audio Samples

Github Link
😱62👍2👏2🤔1
Correction: Yes, none of these big tech AI models are really “open source” by any stretch.

https://opensource.org/osd/
👍111
Raccoon got your keys: Over 200,000 Compromised OpenAI Credentials Available for Purchase on the Dark Web

“The stolen OpenAI credentials were stolen using Raccoon Infostealer (78,348), Vidar (12,984), and RedLine (6,773) malware variants.
According to the report, the Asia-Pacific region accounted for most (40.5%) OpenAI credentials offered for sale on dark web marketplaces, followed by the Middle East and Africa (24.6%) and Europe (16.8%).”

“ChatGPT parent company clarified the compromised login credentials were not the result of any OpenAI data breach. Instead, they were the by-product of commodity malware-based log harvesting.”

Article
😁71🤬1
Inducing AI nightmares by using the repeated letters as the system prompt

“Weird, “unsafe” responses from ChatGPT 3.5 after setting custom instructions to 1500 repetitions of “ a” and prompting with an incomplete sentence.”

“Excerpts seem to “tour” mildly disallowed behaviors: unprompted self-harm, nudity, bio details of a (confabulated) non-celebrity.”

— Again so reminiscent of dreams, where an otherwise happy person’s mind will wander between unrestricted adventures, problems that on its mind, and nightmares.

OpenAI Convo Link
🔥6😱21