Stories featured today from @radioproducer (Yann LeCun interview), @AustinCarr + @ianmking (profile of Nvidia), @JLDastin + @annatonger (Google warns against putting sensitive info in Bard), @ashleyrgold + @AndrewSolender (First look at bill denying section 230 for AI)
GENERATIVE AI
-
Falcon’s Rise Signals New LLM Competitors on Horizon
By
–
The challenges it faces go beyond just those that go into making a commercially permissive model. If Falcon came out of nowhere and was highly competitive on some benchmarks, what else is in the works that can do the same—or pass—LLaMA? Unclear! But surely they are in development
-
Meta dominates AI landscape with PyTorch and LLaMA models
By
–
Meta has consistently outmaneuvered Google (and OpenAI/Microsoft by extension) in AI, both open source and the sheer volume of research it releases on a regular basis. It's pretty much locked up the ML dev ecosystem with PyTorch, and is well on its way w/ LLaMA.
-
Emerging Open Source Language Models After LLaMA Leak
By
–
But several other models have emerged since LLaMA leaked in early March. MPT-7B from MosaicML. RedPajamas-INCITE from Together (based on a data set similar to LLaMA). Alpaca. Vicuna. Instruction-tuned versions of all of the above. Plus *many* domain specific ones like Starcoder.
-
Falcon 40B Apache release may accelerate Meta’s LLM timeline
By
–
People I've talked with have speculated that the release—and abrupt change to an Apache license—of Falcon 40B might accelerate Meta's timetable for release. It made an enormous splash as being both commercially permissive and competitive with LLaMA.
-
Meta evaluates commercial release strategy for 3B AI model
By
–
I reported earlier last month that Meta was evaluating what model(s) to make commercially permissive, and at the time was evaluating a 3B model. Obviously a lot goes into putting out a commercially permissive one vs. just one for research.
-
Meta Makes LLaMA Commercially Permissive Amid Open Source LLM Race
By
–
Today's issue is about the race for open source LLMs that seems to speed up literally every week. This week @theinformation reported that Meta is working on making its LLaMA model commercially permissive. Meta's been evaluating this since as early as May.
-
Voicebox: Generative AI Speech Model Advancement
By
–
This work represents an important step forward for this field of research. Just as other work has done for images + text, we believe the generalization & scalability of Voicebox could usher in a new era of generative AI for speech — we're excited to continue pushing it forward.
-
Voicebox: Generative AI for High-Quality Audio Creation and Editing
By
–
Like other generative AI work, Voicebox is able to create high-quality outputs from scratch or modify samples, but instead of images/video, it produces high-quality audio. Unlike autoregressive models, it can modify any part of a given sample — not just the end of a clip.
-
Voicebox Outperforms Specialized AI Models with In-Context Learning
By
–
Voicebox generalizes across tasks and outperforms single-purpose AI models through in-context learning. It beats the previous state of the art in a number of ways. Better word error rates Improved audio similarity Up to 20x faster performance