Google blows it! Was it “Woke Gone Wild”?

When a giant stumbles, it hits the ground hard.  Google, who literally invented the Transformer architecture that underlies modern AI, last week shot itself in the foot - in fact practically blew its foot clean off - when its flagship LLM, Gemini, was caught generating absurdly “diverse” images of historical figures - such as a Black Founding Father. (See below for more absurd examples.) A furor erupted, with Google being accused of pushing “diversity” in its image generation to the point of “Woke Gone Wild.” Google immediately shut down Gemini’s image generation for retooling. Google stock tanked, erasing $70 billion in shareholder value in 24 hours. And there are even rumors that Google CEO Sundar Pichai may lose his job.

The uproar seems clearly an over-reaction to what is likely a fairly simple glitch in Gemini’s system prompt which should be quickly fixed. But the underlying lesson here - other than racists gonna racist - is that Google is in a trap of its own making. It invented the architecture of modern AI, then let a scrappy upstart like OpenAI steal its thunder. It has been slow to respond with a better AI model, or even an equally good one, all the while promising how good it’s going to be. OpenAI has had its own WTF? moments - remember when the Board blew up? - but as the newcomer they have been given some slack. Google now faces extremely high expectations, and any slip will be taken as evidence that it has lost its touch. Only time will tell whether Google implodes like Kodak, or comes roaring back like Apple after Steve Jobs returned. In either scenario, it seems increasingly likely that Sundar will be spending more time with his family.

Google’s stock tanks after Woke Gone Wild fiasco

Robot startup Figure AI gets $675 million from tech titans

Humanoid robot startup Figure AI has just snagged $675 million from a Who’s Who of tech industry icons, including Jeff Bezos and Nvidia. The deal values the two-year-old startup at $2 billion. Figure uses AI to make its robots intelligent, and has already signed a deal with BMW to put the robots into the carmaker’s plant in South Carolina. Humanoid robots are coming soon, to a factory near you.

Figure AI snags $675 million from Bezos, Nvidia, and other tech stars 

Mistral is Europe’s answer to ChatGPT, and it’s on a roll

Mistral was founded last April by a trio of young AI superstars in Paris, and backed by a bevy of deep-pocketed power brokers hoping to make the company the OpenAI of the EU. Since then, the trio have released one banger of an open source AI model after another (the first one a mere 3 months after getting funding) each one bigger and better than the last. And all the while, the 3 founders exuded a supercool Eurotrash-meets-cyberpunk swagger, as only the French can.

This week they released their newest version, called Mistral Large, which is almost as capable as the industry-leading GPT-4 from OpenAI, putting them in very elite company indeed.

Mistral’s newest model is almost as capable as OpenAI’s GPT-4

The company also has publicly released a ChatGPT-like chatbot they call Le Chat (naturally…). Sign up for the beta version at this link:

Le Chat signup

This week, Mistral announced not one but 2 strategic partnerships with US tech giants.

Microsoft signs a multi-year pact with Mistral to use its models, and invests in the company at a $2 billion valuation.

Microsoft invests in Mistral

And finally, Amazon announces that Mistral will be included in its Bedrock AI platform on Amazon’s industry-leading AWS cloud service.

Amazon includes Mistral in its AI offerings

Not bad for 10 months’ work.

Swedish FinTech firm replaces 700 customer service reps with AI

Klarna, a Swedish payment processing firm, recently announced that they had installed an AI chatbot for customer service requests. The chatbot, based on OpenAI models, ended up handling 2/3 of all customer queries, or over 2.3 million requests. Resolution time declined from an average of 11 minutes down to 2 minutes, accuracy was greater, and customer satisfaction was equal to that with human agents.

Swedish payment firm replaces 700 customer service reps with AI

Klarna’s CEO took to X/Twitter to brag about the results, but also to warn of the potential of AI to displace lots of customer service workers in the very near future. (There were no associated layoffs at Klarna - they had already outsourced customer service.) The switch to AI will mean an extra $40 million in profit this year for the company.

Klarna CEO warns of possible huge impact on customer service jobs

AI learns to control the plasma in a fusion reactor

To harness the limitless power of fusion, scientists have to learn how to control a superhot hydrogen gas called a plasma. Plasmas in fusion reactors are contained by superstrong magnets, and the interaction between the plasma and the magnetic field is inherently unstable, requiring constant split-second adjustments. Scientists have now developed an AI that is faster and better at containing plasmas in a magnetic “bottle’ than any human or machine before it. Practical fusion power is incrementally closer.

AI learns to control the plasma in a fusion reactor

Nvidia CEO says nobody should learn to code any more

Jensen Huang, the tattooed, leather-jacketed CEO of dominant AI chipmaker Nvidia, says that in the age of AI, learning to code is a waste of time. We should put all our efforts into inventing computers that communicate with us in human languages, not computer languages. (Which, not-so-coincidentally would mean a lot more demand for Nvidia’s chips.)

Nvidia CEO says nobody should learn to code any more

Tumblr and WordPress negotiate to sell user posts to train AI

Last week it was Reddit that was strip-mining its users’ posts to sell to AI companies for training data. This week Tumblr and WordPress are reportedly negotiating with AI companies for payment for their text and image data. Pushback from users is growing over having their posts monetized in this way without their consent.

Tumblr and Wordpress negotiate to sell user posts to train AI

AI researchers: Human-level AI is closer than we thought

A survey of over 2,000 researchers who have published in top-tier AI journals indicate that, on average, over the past year these highly informed AI scientists have significantly revised downward their predictions about how long it will be before AI achieves various human capabilities. In aggregate, they predict that the Singularity/AI-pocalypse/Rapture of the Nerds (when AI is better than humans at everything) has a 50% chance of happening by 2047, which 13 years closer than they thought a year ago.

AI researchers now believe human-level AI is much closer than they thought

Here’s the source paper

Athenahealth survey of 1000 doctors shows AI optimism

In a recent survey of 1000 physician by healthcare informatics company Athenahealth, 90% of physicians report having feelings of burnout on a regular basis, and 64% say that they are overwhelmed by clerical requirements. On the brighter side, 84% say that they feel optimistic that AI can help this situation.

90% of doctors report burnout; 84% think AI can help

AI predicts mortality in dementia patients

Researchers at the Icahn School of Medicine at Mount Sinai MIT have developed an AI model to identify key predictors of mortality in patients with dementia of various types. Surprisingly, the study found that neuropsychological test results were better predictors of mortality in dementia patients than their age-related conditions such as heart disease and cancer.

AI predicts survival in dementia patients

AI unlocks secrets of rare diseases

Rare diseases are hard to study due to the inherently limited sample sizes, but in the aggregate, the totality of all rare diseases affects a considerable number of people worldwide. Now the Barcelona Supercomputing Center, which we have reported on in the past, has developed a method using AI to correlate the data from small numbers of patients with a given rare disease with huge biomedical databases to better understand the disease processes in the condition. Most recently, data from 20 patients with myasthenic-congenital diseases in Bulgaria was correlated with biomedical databases of genes and their physiological effects. This allowed a greater understanding of why patients with the same gene mutation can have different severities of illness.

AI unlocks secrets of rare diseases

AI helps Nebraska hospital cut nurse turnover in half

Omaha-based Nebraska Health, which operates 2 hospitals and over 40 outpatient clinics, cut nursing turnover in half when it implemented an AI-powered task management system for nurse managers, which allowed these frontline nurse leaders to better manage staffing issues.

AI helps Nebraska hospital cut nurse turnover in half