🔥AI Hot Takes: AI Extinction Warning; AI Discovers a New Antibiotic & Turns Brain Scans into Videos

👋🏽 Hey friends,

Before I get to my Hot Takes on the most interesting AI developments of the past week…

…It’s good to see you all! I’m back after a brief hiatus. I spent most of April in Japan, Thailand and Singapore with the fiancé. I actually used A.I. to plan 95% of the trip. I trained an instance of ChatGPT to suggest places to visit, foods to try, and activities to embark upon — all personalized to Deborah and me. I have a note on how I did it in the “Updates” section of the newsletter.

I’ve been working on a few things since I got back — a curriculum for a live course on Coworking with A.I. (how to maximize your use of AI, find AI tools, optimize your org with AI) with AI expert and my co-founder Matt Schlicht (use EARLY20 for 20% off); a new AI column for The Information (check out my most recent column on AI and Jobs); and a secret project or two. I also just gave a keynote about the future of A.I. on a table to 200+ entrepreneurs and investors in SF. You can read a recap on my LinkedIn. Thanks Jeremiah for hosting the event and inviting me to speak.

Finally, I’m working on a major update to this newsletter. As part of this change, this newsletter is now called “The A.I. Analyst”. The mission hasn’t changed — I will still analyze the impact of technology, specifically AI, on our society. But there will be a few other changes. (I’m looking to expand; email me if you want to write more about AI). More details in the Updates section at the bottom of the newsletter.

First though, the Hot Takes!

Ben’s A.I. Hot Takes

1. ⭐️HOT OFF THE PRESSES: A.I. Poses ‘Risk of Extinction,’ Industry Leaders Warn. From NYT: “‘Mitigating the risk of extinction from A.I. should be a global priority alongside other societal-scale risks, such as pandemics and nuclear war,’ reads a one-sentence statement released by the Center for AI Safety, a nonprofit organization.” 350 top AI experts, including the CEOs/execs of Open AI, Google Deepmind, Anthropic, and the “Godfather of AI” Geoffrey Hinton, all signed the letter.

🌶️Ben’s Hot Take: That’s a SPICY way to kick off the week! Some of the people closest to AI are saying that we have to treat AI like nuclear weapons. This would, of course, create a lot more regulation of AI — which could help big players like Open AI and Google by preventing new competitors from entering the AI race. I do believe these leaders think AI will surpass us humans very soon (if it hasn’t secretly already) and could pose a risk to humanity without safeguards. That said, I don’t know if this kind of doomer-ism talk really helps the conversation all that much. (I’m not in the doomer camp.)

I will have a deeper column on this topic soon. This requires a much deeper analysis.

2. 79% of American workers believe AI will threaten their salary and lead to pay cuts. 74% think layoffs will happen due to AI. 79% are feeling pressure in work to learn new AI tools. And 57% said they’d take a pay cut if AI meant they got a 4-day workweek. 3,000 Americans across all age groups were surveyed.

🔥Ben’s Hot Take: Workers should be scared. Everyone is underestimating the impact AI will have on the job market. That’s why I argued AI will destroy jobs and that we need a contingency plan in my latest column for The Information.

3. Here’s What Happens When Your Lawyer Uses ChatGPT: He has to beg the court for mercy. According to NYT, a lawyer representing a man who sued an airline relied on ChatGPT to help prepare a court filing. It backfired: it hallucinated fake cases and the lawyer begged the judge for forgiveness.

☄️Ben’s Hot Take: Generative AI is great for structured tasks with rules — e.g. writing a simple contract — but unreliable when the accuracy has to be 100%. Someone is going to make huge bank creating a company that can fact-check GPT-4 and the outputs of other LLMs in realtime.

4. AI Reconstructs 'High-Quality' Video Directly from Brain Readings in Study. Researchers in Singapore were able to use AI to “reconstruct” what someone saw by “reading” their brain waves! The researchers leveraged the Stable Diffusion AI model and fMRI data to create “MinD-Video” — which generates images and video from brain readings.

🔥Ben’s Hot Take: Holy shit! THIS is the kind of advancement that will lead us to brain-computer interfaces (BCI). This is a very fascinating use of a generative AI model (stable diffusion). Just imagine if you could draw from just your mind.

5. Scientists use AI to discover new antibiotic to treat deadly superbug. AI was used by a team at MIT and McMaster University to discover a new antibiotic for Acinetobacter baumannii, which is considered a “critical” threat by the WHO. They trained a neural network with a dataset of 7500 molecules and compared to those that inhibited the growth of the pathogen and discovered the antibiotic, “abaucin”. The full paper is worth a read.

❤️‍🔥Ben’s Hot Take: When you want to convince someone that AI can save the world, send them this. This is the promise of AI doing good for humanity. I’ve been hearing of startups building customized treatments using generative AI and neural networks. I want more of this.

6. Meet Voyager: An Autonomous Agent That Can Play Minecraft With GPT4. It can play Minecraft faster and better than human can. It can find 3.3x more rare goods, travel 2.3x as far, and reach in-game milestones in Minecraft 15.3x faster.

🔥Ben’s Hot Take: Now we are starting to see autonomous agents that are better than humans in very specific areas. If you’re not familiar with Minecraft, I recommend checking it out so you can see how hard it is. (I’ve happily dumped many happy hours building crazy mansions in Minecraft.) The fact that a large language model could be trained to play this game autonomously without constantly dying is huge. If you are unfamiliar with autonomous agents, I strongly suggest you read the Complete Beginner’s Guide to Autonomous Agents by my co-founder Matt Schlicht.

7. Nvidia’s stock price goes to the Moon, thanks to A.I. There is no bigger winner in the AI world right now than NVIDIA, whose stock shot up by 25% after it announced killer earnings and surpassed analyst expectations for its outlook/projections by 50%.

🔥Ben’s Hot Take: You always want to provide the picks and shovels for any industry. Open AI provides the LLM most startups are using (GPT-4), but Nvidia is making the chips that power generative AI. What a crazy rise — from beloved GPU for gamers to the 6th most valuable company in the world! (Bigger than Tesla, Meta, LVMH and Berkshire Hathaway)

8. OpenAI warns over split with Europe as regulation advances. The EU has proposed an “AI Act” for longer than ChatGPT has existed, but if it went into effect as written, it would potentially move OpenAI, Google and other LLM makers to not offer tools like ChatGPT in Europe. Misinformation, copyrights and the role of AI in making decisions are all addressed by it.

🔥Ben’s Hot Take: The AI Act today is too cumbersome and unrealistic for today’s AI companies. But I suspect the language will become more forgiving and it will be like GDPR — a defining technology privacy law that ultimately didn’t hamper innovation.

Updates

1. The A.I. Analyst is looking for an additional, part-time writer to assist me in doubling down on the analysis of A.I. news and world events! AI is moving at light speed, so I’m looking to expand this newsletter! I am seeking someone who can help me find and organize all the interesting AI news stories, developments and tweets and help me upgrade this newsletter into the most helpful & useful resource for AI analysis and insights on the market. If that might be you, send me an email ASAP.

2. If you’re wondering about where I’m taking The A.I. Analyst — my goal is to make this your top resource for AI analysis. There are plenty of great newsletters about AI news, but very few breaking down top stories and what they might mean for both the AI industry and society at large. In the near future, expect 1-2x newsletters a week — one email with my 🔥Hot Takes on top AI news and developments, and occasional deeper analysis of an important topic in AI. This will sometimes include my columns on The Information, as I know some of you are not yet subscribers.

3. My co-founder Matt Schlicht and I are teaching a live course on “Coworking With AI”, available now on Maven. The first class starts June 12th.

   Hundreds of people have asked my co-founder Matt Schlicht and I to help them understand AI, what tools to use, and where it’s going. So we are going to share everything we’ve learned from our 7+ years in A.I.

   If you use the code EARLY20, you will get 20% off the course! Only the first 20 will be able to use it.

   

   There will be some special guest speakers for our Coworking with AI course!

   A.I. is the future, but we know many need guidance on how to best use AI, what AI tools to use, and how to better use AI automations for operational efficiency. We will cover all of it in our two-week course.

4. I spent a lot of time in April exploring Japan, Thailand and Singapore with Deborah. We were reminded of the beauty of this world. I am posting photos from the trip on my Instagram if you want to see our journey to the shrines of Nikko, the landmarks of Kyoto, or the ryokans of Hakone.

   On the A.I. side, I found ChatGPT to be a very competent, very helpful tour guide and travel assistant. I created a specific instance of ChatGPT and trained it on some of our parameters (e.g. no amusement parks, highlight cultural landmarks, find us sculpture art, etc). ChatGPT was a constant and welcome companion on our journey.

   

   I add emojis to the ChatGPT instances I have spent time training, so that I can find them quickly and easily.

   The only real limitation ChatGPT had was its knowledge cutoff date. But I suspect that won’t be a problem as ChatGPT gains more realtime information from Bing and the Browser plugin.

4. This newsletter will continue to evolve in the coming weeks. Expect me to add some more helpful information around AI tools (a top request) and answers to your most-asked AI questions.

   What would you like me to talk more about on The A.I. Analyst? Have a question you want me to answer around AI? Send me an email.

The next newsletter will drop on Friday.

Cheers,

~ Ben