Generative AI News - New ChatGPT Features, HuggingChat, Google, Deepfakes, and More - Voicebot Podcast Ep 317

The Generative AI News (GAIN) rundown for April 27, 2023, is
here. Another week of breaking news has piled up, and we have a
breakdown of the top stories and what they mean for the industry.
The developments include news from ChatGPT, HuggingFace, Google,
Nvidia, Sensory, Hour One, D-ID, deepfake musicians, and more.


Your hosts today are Bret Kinsella and Voicebot.ai's Eric
Schwartz. The top stories in generative AI land this week
include:
ChatGPT En Fuego



Plugging in a new vision: Greg Brockman from
OpenAI demonstrated some new ChatGPT plugin features; several
are jaw-dropping.




The “super app” virtual assistant we were
promised: Brockman’s demo and the discussion about
the product philosophy offer an insight into where ChatGPT is
headed. Move over, Alexa. Get out of the way, Siri. ChatGPT
may be the virtual assistant we have always wanted.




ChatGPT is anything but
incognito: While everything ChatGPT seems to
play out in the public eye, OpenAI recognized that not every
user wanted every one of their chat conversations saved in
perpetuity and used for future model
training. Incognito (i.e., private chatting) is now
available, and a “business mode” is coming soon.


HuggingChat Embraces Open Source



Open source competition for ChatGPT: Hugging
Face stepped up and provided a ChatGPT
alternative built on open source models and data. It’s a
smaller AI model than ChatGPT and is pretty good.


Deepfake Entertainment



Drake, The Weeknd, Bad Bunny, and Rihanna go
viral: Viral hits from big stars are
common. Deepfake viral hits mimicking the voice,
style, and likeness of big stars may also become common.
ghostwrider777 strikes again!




Joe Rogan comments run deep: New deepfakes
mimicking Joe Rogan’s podcast have the comedian and
commentator talking about a “slippery” slope.




Grimes jumps on board: The musical artist
says she will split royalties 50/50 with anyone
deepfaking her voice. She has no label and no binding legal
constraints giving her more flexibility than most musicians.


More Virtual Human Expansion



Prompt-to-video: Hour
One introduced a new text-to-video solution that enables full
video generation for presentations from a single prompt.




Canva gets digital people: D-ID
introduced a new Canva app that enables you to add
generative videos to any project.




Chatbots are suddenly
popular: Character AI landed $150M in
funding at an obscene valuation. Virtual Elon Musk, Mark
Zuckerberg, and 2.7 million other chatting avatars with
personalities have driven 100M user visits in just two
months.


Google Ups Generative Game



Bard learns to code: Google is slowly
catching up with the generative AI leaders. It’s ChatGPT
competitor—or, is it a Bing Chat competitor—can now code.
This is not a true competitor to GitHub Copilot yet.




Sec-PaLM gets into security: Google also
rolled out a new cybersecurity solution with the parsimonious
name of Google Cloud Security AI Workbench. It is based
on a fine-tuned version of the PaLM large language model
(LLM).


Nvidia and Sensory Plug Market Gaps



ChatGPT gets an edge: Sensory rolled out a
new hybrid on-device and cloud solution that can enable the
use of ChatGPT and similar services on devices.




Nvidia on rails: NeMo, Nvidia’s LLM, now has
a new feature for adding guardrails to other LLMs to align
model outputs with companies’ safety and security
requirements. NeMo Guardrails is open source and
designed to work with any LLM.