Welcome to this week's roundup of AI news optimized for human audiences.
This week the AI tried way too hard to be woken up.
New AI models are popping up in every single place and so they want your data.
And an $800 million film studio project was canceled due to Sora.
Let's dive in.
Oh, what a tangled web we weave…
As generative AI systems grow to be cobbled together and more obscure, they accumulate a pile of “technical debt.” New studies explain how patchwork AI development sets things on a path that inevitably ends in disaster, and that it's unimaginable to repair AI hallucinations.
Google's Gemini went into overdrive this week in dramatic evidence of the impact technical debt can have. Google has shut down Gemini's image generator feature after the corporate was criticized for generating absurd and inaccurate images of historical figures.
Apparently Gemini really doesn't like drawing white people.
Google further expanded its collection of confusing AI product names when it announced Gemma, a collection of lightweight open LLMs optimized to run locally on NVIDIA RTX PCs.
These little LLMs are literally pretty impressive and the flexibility to run them locally might be game-changing.
The imitation game
OpenAI criticized The New York Times, claiming the newspaper paid someone to “hack” ChatGPT to provide evidence for its copyright lawsuit. OpenAI's lawyers are principally saying, “Yes, ChatGPT can output verbatim NYT content, but only in case you try really hard.”
It's not only the NYT that doesn't want its content for use as AI training material. A brand new study shows that major news sites are increasingly blocking AI web crawlers. Is it a journalistic principle or are they waiting for offers of money?
User-generated content is becoming a giant goal for AI corporations. OpenAI and MidJourney are actually attempting to buy your WordPress and Tumblr data to coach their models.
The company behind the deal is throwing around terms like “attribution, opt-outs and control,” but have a look at the way it worked out for the NYT.
The argument that OpenAI's models don't directly copy things from training data just got a bit harder. A Copyleaks report found that about 60% of GPT 3.5 editions are plagiarized.
So that probably implies that around 60% of scholars’ homework can also be plagiarized. Does anyone even go to a library for books anymore?
Big money
The sum of money invested in AI is staggering and Nvidia appears to be a firm favorite amongst investors. The recent rise in share price makes the corporate larger than Google or Amazon, with a valuation of $2 trillion.
Lots of money is flowing into projects to develop advanced, universally applicable humanoid robots. Robotics company Figure AI closes a $675 million funding round and is backed by some big names within the AI industry.
It would require plenty of energy to maintain these GPUs and robots running. Clean energy from nuclear fusion could grow to be a reality prior to we thought, and AI has a very important role to play.
AI celebrities
There's a surefire option to grow to be a millionaire in only three months, and British star Ben Fogle wants to indicate you do it before the banks close with this straightforward trick.
However, this version of Ben Fogle is just certainly one of many AI deepfakes that scammers have posted on Facebook. How do these ads even manage to pass Meta’s ad review process?
Some real celebrities are also feeling the AI crisis. Actor and filmmaker Tyler Perry is putting the brakes on his $800 million studio expansion. After seeing how easily Sora generates video content, he realizes that investing in studios, sets, and filming equipment might be not such an excellent idea.
If Perry saw how easily ElevenLabs added AI-generated audio to the Sora videos, then his project will certainly not occur.
We were overwhelmed by the Sora announcement, but felt something was needed…
What in case you could describe a sound and create it with AI? pic.twitter.com/HcUxQ7Wndg
– ElevenLabs (@elevenlabsio) February 18, 2024
The real stars of the AI show are a number of the super smart women developing the technology. If you're bored with all the time hearing concerning the men behind AI, try this list of the highest women making big changes in AI.
eye see
The researchers wanted to seek out out whether GPT-4 might be useful in answering questions and cases from glaucoma and retina patients. How do you think that it compares to skilled eye care professionals? You will want to reconsider a profession in ophthalmology.
Another model making AI headlines this week got here from French startup Mistral AI. The company will not be even a yr old and has just released a brand new chatbot and model that’s pretty near the performance of GPT-4. API prices could pose an actual threat to OpenAI.
What we really need is an AI assistant that may interact with multiple software tools to do the tasks we don't wish to do. Meta and UCSD have brought us one step closer with ToolVerifier, which helps LLMs select the fitting tools once you task them with completing a task.
In other news…
Here are another clickable AI stories we liked this week:
- Stability AI released Stable diffusion 3 and it's really good at adding text to the pictures it produces.
- Grok (not Grok) has a dedicated hardware platform that processes AI model inference much faster than other cloud platforms.
- Microsoft's take care of French startup Mistral AI is under EU review.
- The people behind the fake Biden robocall are a Democratic consultant work for a competing candidate and a Street magician.
- Samsung is latest High bandwidth storage is 50% higher than its previous chip and will speed up AI training by 34%.
- Phrase your prompts as Scene from Star Trek can improve the response of an LLM.
And that's a wrap.
There's plenty of buzz surrounding Google's AI projects, but currently they've been making headlines for all of the flawed reasons. Did you get super woke images when using Gemini? Can you tell Gemini from Gemma and Pro from Ultra?
You can have joined Team OpenAI within the NYT saga, but how do you are feeling now that AI corporations want your Reddit, WordPress and Tumblr content? Take my data and do something good, or hands off?
If the industry can address social awareness and copyright issues, are you able to imagine what the industry could achieve with a brand new focus? I've added nuclear fusion energy and fully immersive VR worlds with multi-sensory feedback to my wish list. Simply.
Share with us your boldest AI predictions for 2024 and please send us links to interesting AI stories we can have missed.
