I delayed sending this until Friday in hopes that the AI Act negotiations would conclude in time for this week's update but, alas, no deal (yet). There's still time, though, and still lots of news, as well.
- The EU is so, so close on the AI Act, having made incremental
after incremental
after incremental
advances. The New York Times published a great synopsis of how the EU got to where it is with the AI Act
. Meanwhile, the U.S. Senate is getting ready for a bill-a-palooza in 2024
. If you want to focus on what’s already in flight–i.e., the Biden AI EO–Stanford has a good guide to it
.?
- Good! More and more companies – from Walmart
to Sanofi
– are talking about their (relatively similar) approaches to responsible AI (let’s hope they walk the walk, too).
- The open AI community
(led by…Meta, Intel, IBM, etc.?) is organizing into an “AI Alliance
” to promote its values.
- The New York Times published a list of AI luminaries
and apparently forgot about women
, and Asians
, and many others. But, hey, the NYT at least had a good story about a campfire conversation in 2015
.
- People are talking about AI transforming the world of work, but what does that look like at the ground level
? For one, it might mean that AI levels the playing field
. For small businesses, it’s still an area of exploration and growth
. For junior consultants, could it be a rocket ship to promotion
? And for programmers, it’s changing everything
.
- Speaking of the working world, Microsoft’s Copilot is now one year old
, it’s showing up with new upgrades
, and evidence is suggesting that it has the potential to significantly improve productivity
.
- Using black-box AI for determining creditworthiness
scores is–to put it mildly–problematic and it will lead to legal challenges
.
- ChatGPT ‘won’ Wikipedia
in 2023.
- Arvind Narayanan, Sayash Kapoor, and Seth Lazar point out that RLHF is intended to counter accidental harms
, not intentional ones, and shouldn’t be seen as a panacea for all LLM-related risks.
- Google announced
Gemini
which, in internal testing, achieves parity with GPT-4. More here
and here
. Early reviews are somewhat
mixed. Sundar Pichai spoke with MIT Tech Review
about it.
- Sure, these cool new AI video tools could be used for TikTok videos
(as demoed) but they’ll probably be used for deepfakes of politicians and famous people shortly thereafter.
- “Be audit you can be?” The IAAA
came into being as a new, AI audit certification organization
.
- The SEC is weighing in on “AI washing,”
or using unfounded AI hype to improve stock prices.
- The OMB started to receive comments in its AI rulemaking process. Here are CIPL’s
.
- Erik Brynjolfsson and Gabriel Unger wrote an article for the IMF
about the macroeconomic implications of AI in the economy.
- The Responsible AI Institute released guidance tools
for addressing LLM hallucinations, corporate policies, and vendor assessments.
- Nirit Weiss-Blatt published a deep-dive
into the connection between the EA movement and concerns about AI as an existential threat.?
- Relatedly, Forbes covered Based Beff Jezos and the accel/decel debate
. Bloomberg also covers the increasingly religious-like debate
. And Ben Thompson dives into what accelerationism means for the Internet and content
(but, as Adam Singer pointed out this week, fighting remix culture and AI is probably a losing battle
).
- Stability didn’t get the reprieve it was hoping for in the UK courts
.
- But Mistral appears to be getting the investors
it is hoping for!
- When they say there can be a GPT for everything, I don’t think this is what they meant
.
- IAPP and EY published a report on the professionalization of AI governance
.
- Not too surprising: Sam Altman was named Time’s CEO of the year.
?
- Meta says that everything is hunky dory for them and their open approach to AI models
.
- In case you’re wondering where the market is going, AI startups are continuing to raise >$100m Series As
, Musk’s xAI is raising up to $1 billion
, and an AI infrastructure startup just hit a $9 billion valuation
.
- Apparently, Amazon’s Q model for enterprise use has some shortcomings
.
- Last week we learned that Sports Illustrated used AI writers to use AI to write content. This week, some executives paid the price
.
- Meta’s AI team is on a roll, releasing their “Purple Llama
” trust framework.
- The U.S. Patent Public Advisory Committee advised the USPTO
on how it might approach various questions relating to AI.
- People have been speaking more of sovereign AI
recently. Pablo Chavez explains what that means
.
- There was a lot of fuss earlier in the week about how test scores are falling. Brown’s Zelma
tool can help parents compare their kids’ schools with other districts around the country.
- Research shows that AI can improve breast cancer detection rates
.
- One way to avoid researchers flagging disinformation on your platform is to combat and minimize disinformation on your platform. Or you can donate lots of money to the researchers’ institution and the research might stop
.
- The New Yorker published a long article on Microsoft and OpenAI
. It’s well worth the read.
- Some seem to think that the changes at OpenAI might lead to greater competition
in the AI space.
- The U.S. GAO is reportedly releasing a report
soon on federal agencies’ use of AI.
- The geo-political fight over chips continues: the U.S. reportedly forced Saudi investors out of a U.S.-based chip startup
.
- Omidyar Network is investing to support inclusive AI development
.
- Federal agencies are starting to come up with rules and regulations to implement the Biden Administration’s EO and companies are starting to prepare for what those rules and regulations might look like
.
- Some are trying to slow down AI development. And some are trying to speed it up. And then there are some aiming for maximal acceleration
.
- Vitalik Buterin published a very lengthy and interesting manifesto
for the acceleration of AI development.
- General Mark Milley highlights the need for AI investment for U.S. defense
in a white paper he recently published.
- Axios makes a good point: Deepmind’s discovery of new materials might help reduce the reliance of the world on the geographic happenstance of rare earth minerals
.
- Runway pulled Getty Images
into their orbit.
- It would be wild if AI + lasers make stethoscopes obsolete
. And AI as an oncology service
continues to expand.
- Not unrelatedly, here’s a good argument about the need for clear guidance and regulation of AI deployments in the medical field
. Many doctors increasingly think some sort of guidance is important
.
- As Axios points out, AI uses lots of energy
. But I think an important question is: a lot of energy…as compared to what
?
- Yann LeCun seemingly loves to play the contrarian
. But he and Meta may have a good point regarding the capabilities of many LLMs
. In any event, the IEEE is already asking if the Turing Test is past its prime
.
- Researchers ask: are open models catching up with closed models
?
- MIT is studying how people perceive AI-created content
.
- NYU Law held a symposium
related to AI.
- If we end up regulating AI appropriately, it might end up being our greatest. tool. ever
.
- AI might help in the fight against widespread fentanyl distribution
.
- Deepfakes don’t hurt only those fooled by the content; they can also be devastating
to those who are deepfaked.
- O’Reilly published a good glimpse into the ways in which enterprises are (and are not) taking advantage of AI advances
.
- The Bank of England is, like other banking authorities, engaging in an evaluation of the risks AI developments may pose to financial institutions/markets
. Maybe some of the risk is coming from news like this: McKinsey thinks banks can profit to the tune of $340B annually through increased use of AI
.
- A Brazilian lawmaker used ChatGPT to write a law
(and to make a point in doing so).
- Amazon reportedly fired producer Joel Silver over his hesitancy regarding the use of AI to complete projects
during the recent Hollywood strikes.
- Semianalysis explains the economics behind cloud GPUs
.
- Rain
is hoping to become the next Nvidia based on brain-inspired chips. Sam Altman apparently has a stake and reportedly got OpenAI to buy from them
.
- Fascinating: Singapore is <1% of global GDP but accounts for around 15% of Nvidia’s revenue
. Meanwhile, Nvidia is also big in Japan
. And they’re working to ensure they can sell into China while complying with U.S. export restrictions
. And the UAE is moving away from Chinese-developed hardware to maintain access to U.S.-developed chips
.
- Speaking of, Huawei apparently caught people off guard
with the sophistication of their recent chip release.?
- And, on the topic of chips, it looks like AMD is going to fight hard to claim market share
.
- Concern over the development of AI in China is helping to set the regulatory path, at least in the U.S. Senate
.
- Bob Rosin at Defy.vc
wrote about the next phase of AI development: solving hard tasks
.
- Senegal is setting a national AI strategy
, following on Rwanda’s example
. Meanwhile, Kenya’s tech industry is pushing back
on proposed regulations.
- Clearview AI is in the process of settling
the consumer class action against them.
- AI-powered drive-throughs aren’t working so well
.
- Funny: A16Z’s big ideas for crypto in 2024
generally try to tie it to AI.
- Of course there’s an AI Santa
.
Data, Privacy, Risk & AI Governance
11 个月This is an amazing source of information!