For instance, re the use of water & LLMs – imagine the joy & innovation that will come from any AI company reporting how their plans for new data centers were created via community engagement including indigenous populations to ensure water is protected for generations to come.
GENERATIVE AI
-
Web Grounding Reduces LLM Hallucinations in Knowledge Assistant
By
–
Model hallucinations are a big concern for organizations adopting LLM-based systems. This is why we’ve introduced a web grounding for Coral, our knowledge assistant. Watch NLP expert @itsSandraKublik demonstrate how it works. pic.twitter.com/4JLdMJEG7H
— Cohere (@cohere) 24 novembre 2023Model hallucinations are a big concern for organizations adopting LLM-based systems. This is why we’ve introduced a web grounding for Coral, our knowledge assistant. Watch NLP expert @itsSandraKublik demonstrate how it works.
-
Distil-Whisper: Open-Source State-of-the-Art Speech Recognition Model
By
–
Distil-Whisper is a speech recognition model with state-of-the-art results for transcribing any kind of audio. Plus, it is completely open-source and you can use it right now to try it on their huggingface space.
-
AI’s Lightning-Fast Computational Power Drives Business Success
By
–
The thing I like most about this, aside from seeing a lot of businesses having a successful day of course, is very subtly highlighting "We can add numbers together really, really quickly."
-
Model Hallucination Problem Will Be Solved Eventually
By
–
I am not saying model hallucination doesn't happen or that it's not a problem. But it will be solved one day.
-
AI Diagnostic Capabilities: Recognizing Limitations and Reliability
By
–
I know it is telling me what I want to hear, I know it probably doesn't actually have the capability of diagnosing this yet rigorously, but yet, the following end note was striking:
-
Using ChatGPT Prompts for Effective Copy Editing Tasks
By
–
And as long as I'm live tweeting conversations with computer programs, the following prompt worked decently well to get ChatGPT to simple copy editing. (10 of 15 issues identified were real and 3 I would have let an editor convince me of.)
-
DALL-E Refuses to Generate Specific Individual Faces
By
–
Dall-E: No, generic stand-in is the best I can do. I'm happy to suggest other concepts which would result in something close but not representative of the face of a specific individual.
Me: Fair enough! -
Request to Make Fictional Character Resemble Real Criminal
By
–
Hours later: Me: Say, question for you. I don't know if you could do this, but if I were to say this character is obviously a stand-in for a particular real-life criminal mastermind, who is certainly in your data set, can you make it look more like him specifically?
-
DALL-E Image Generation: Refining AI-Generated Character Design
By
–
Me: Make the character in the style of a classic James Bond *villain*, like say Blofeld. Also, make him a middle-aged Chinese man.
Dall-E: Here.
Me: Ooh that is so close but could you put the overhead light fixture back, that was a great touch.
Dall-E: Here.
Me: Perfect.