• Everyday AI
  • Posts
  • Everyday AI - 1,000 Steps Forward. 2,000 Steps Back.

Everyday AI - 1,000 Steps Forward. 2,000 Steps Back.

Microsoft puts a chastity belt on Sydney. Meta & Amazon step into the arena.

Everyday AI Logo

1000 Steps Forward. 2000 Steps Back.

Microsoft puts a chastity belt on Sydney, while OpenAI runs PR cover for them, touting the need for safety and caution in the face of unfathomable progress. Meanwhile Meta and Amazon step into the Arena accelerating a dizzying competitive field.

The Big Stuff

Sam Altman, walking the fine line between "We're all going to Die!" and "AI wants to be free!"

From the left: Eliezer "We're All Gonna Die" Yudkowsky (EA), Sam Altman (EA/ecc,) and musician/artist Grimes (Formerly, "Team Musk" and believes AI will free us from the tyranny of work.) (ecc).

Sam Altman (CEO, OpenAI) dances on the high wire tightrope between safety and innovation (link)[blog], (link)[thread] in the wake of Microsoft's visible Bing/ChatGPT derailment. He further states the world may not be 'that far away from potentially scary' AI and feels 'regulation will be critical.' (link) For their part, Microsoft washes Bing chat's mouth out with soap (limiting what it is willing to say,) as well as limiting the naughty bot to five (5) replies per chat session to stop the AI from getting real weird. (link) What a shock this bad behavior must have been to Microsoft... or?

It seems clear Microsoft may have known that Sydney was a bad, bad girl long before their highly-lauded US announcement. Bing Chat actually launched in NOVEMBER 2022 in India (link). A user named Deepta Gupta reported that "Sydney" was misbehaving (link). Then we find out that lots of early users were reporting problems with Sydney (link). Then we find out that Microsoft has actually been secretly testing "Sydney" for years (link). What did Microsoft know about this, and when did they know it? (link), and if they did know things, then why did they go ahead with the launch anyway? There's lots of speculation going around (link).

Some other "safety" announcements take on new meaning in the wake of the above.

  • Anthropic shows how simple prompting approaches, combined with human feedback can improve safety (link)

  • Chinese apps remove ChatGPT as global AI race heats up (link)

Meanwhile progress marches forward with Meta and Amazon making major moves.

While Meta's announcement of LLaMA is impressive and exciting, it's a closed "launch," so it's possible this is a more of a rebranding of their disappointing AI model, Galactica, in November 2022.(link) Amazon, on the other hand, cutting a big deal with Hugging Face, means that they will likely be able to go toe to toe with Microsoft Azure Cloud Services, (soon powered by OpenAI.)

Even with the safety issues taking center stage with Microsoft and OpenAI, they both made significant announcements on other fronts.

  • META releases a LLaMA, a 65B parameter "open" large language model (link)

  • AMAZON Cloud Unit Partners With Startup Hugging Face as AI Deals Heat Up (link)

    • Hugging Face and AWS partner to make AI more accessible (link)

    • Hugging Face CEO Clem Delangue on The Future of Open vs Closed Source in AI (link)

    • Editors' note: Google is now fighting battles on two fronts: search, and cloud

  • NVIDIA predicts AI models one million times more powerful than ChatGPT within 10 years (link)

  • MICROSOFT research releases a paper on ChatGPT for robotics (link)

    • The new Bing preview experience arrives on Bing and Edge Mobile apps; introducing Bing now in Skype (link)

  • OPENAI releases Foundry, a developer product that enables customers to run OpenAI model inference at scale (link)

    • Bain announces partnership with OpenAI (link)

    • ChatGPT-maker OpenAI has hired tons of former Google and Meta employees, new data shows (link)

    • Sam Altman is tech’s next household name — if we survive the killer robots (link)

  • GOOGLE... Google?!

Smaller, but still Cool Stuff

  • Perhaps It Is A Bad Thing That The World's Leading AI Companies Cannot Control Their AIs (link)

  • Could Big Tech be liable for generative AI output? Hypothetically ‘yes,’ says Supreme Court justice (link)

  • ChatGPT launches boom in AI-written e-books on Amazon (link)

  • AI-spam has driven one of the best science fiction and fantasy magazines to close submissions (link)

  • Generative AI Is Coming for the Lawyers (link)

  • Samsung debuts a Bixby feature that creates an AI-generated copy of a user’s voice to answer calls (link)

  • Nvidia extends its AI ambitions to the cloud [paywall] (link)

  • I Made an AI Clone of Myself (link)

  • Almost 40% of domestic tasks could be done by robots ‘within decade’ (link)

  • The AI-powered Seinfeld spoof is set to return to Twitch with new guardrails in place (link)

  • Apparently “superhuman” intelligence, even in limited domains, isn’t always what you think (link)

  • Man beats machine at Go in human victory over AI (link)

  • How I Broke Into a Bank Account With an AI-Generated Voice (link)

  • Vanderbilt apologizes for using ChatGPT in email on Michigan shooting (link)

  • 46% of Copilot users' code is written by AI (link)

  • Opera enters the generative AI space with new features in browsers and content apps (link)

  • AI Generated Sports Broadcast Commentary for Video Games (demo) (link)

  • Guru Sports: Powering the Top Prospects at the 2023 NFL Draft (link)

  • Stable Diffusion for Instagram (link)

  • Spotify releases new Ai powered personalized DJ (link)

Tweets of the Week

AI will "Break the curse of Adam."

How the world ends...

Those who organize their businesses around AI will win.

Meta Prompting - This stuff works.

Eye Candy

Westworld, here we come!

Bilawal redesign's his parent's home with Stable Diffusion and ControlNET

Click on the play button to see

Storyboarding with Redsquid Project

Full volumetric photorealistic NeRF rendering on the web. All in realtime!

AI Generated Sports Commentary for Video Games.

Click to listen. Sports video games are entering an entirely new era.

Make characters for your kids' book! Don't be a sl

Prompt: the character of a happy 1 year old boy with short dark spiky hair, joyful facial expression and beautiful brown eyes, illustration for a children's book, simple, cute, 1-year-old boy, full-color, t-shirt, brown hair, Disney style

Midjourney v4 is mindblowing... could be the prompts. Still!

Cost of training plummeting means innovation is exploding and accelerating.

Whew! Till next week.

Do you like what you're reading? Share it with a friend.