Collaborative Intelligence Week in Review - 01Mar2024
Best business article(s) I read this week…
This is an article from Elad Gil, a successful tech entrepreneur and investor. In it, he poses several interesting questions about each layer of the current AI technology stack. Two of several thoughts that came to me while reading:
Understanding the evolving AI landscape (particularly LLMs) is crucial for strategic planning. The potential shift towards an oligopoly in this sector, as discussed by Gil, could significantly impact competition, affecting how companies leverage AI in the future.
The role of cloud providers in the AI ecosystem highlights their importance. Cloud platform selection decisions will have a direct impact on an organization’s AI cost structure, its ability to scale, and its agility in a rapidly changing market environment.
Best technical article(s) I read this week…
Direct Preference Optimization
The authors argue that using Reinforcement Learning with Human Feedback (RLHF) is fraught with problems. They propose a new algorithm, which they have named Direct Preference Optimization. They claim this approach is more “stable, performant, and computationally lightweight” than RLHF. One thought I had while reading this:
This approach (assuming what they write is true) simplifies training time and expense while better capturing human preferences. Such improvements should lead to faster development cycle iterations, enabling development teams to more quickly refine and deploy AI solutions that better meet user requirements.
Other item(s) of note…
Datamation’s Top 100 AI Companies
Including this article doesn’t represent an endorsement of any company on the list, but I am passing it along as an FYI. You might find something of interest. I did.
ChatGPT goes temporarily “insane” with unexpected outputs, spooking users
I experienced this problem first-hand. You might have, too. It happened last week, not this week, but I didn’t see this article until Monday, February 26.
AWS Innovate — Free Online Conference
This is the registration link to Amazon AWS’s free Innovate Conference, which is focused on “how you and your organization can leverage the latest advances in generative AI.” It includes enough speakers from other organizations that it may be more than a sales pitch for AWS services.
What do Threads, Mastodon, and hospital records have in common?
This is from January, but I just saw it this week. It describes how hospitals use “federated learning” to train AI systems without violating HIPAA.
Coolest thing(s) I saw…
This was passed along to me. It’s a simple game driven by generative AI whose popularity is growing. You merge two words, and another word that represents the combination pops up. For example, “water” + “earth” might yield “mud.” It’s interesting to see what odd combinations of words produce.
This is the latest research advancement from Ken Goldberg’s team at UC Berkeley. The accompanying video shows two robotic arms suturing a wound in artificial skin with six stitches. The technology is still not ready for prime time, but it is advancing.
Company to monitor…
This is a new section. It will not appear every week. It aims to bring more attention to companies I find interesting.
Groq is a company specializing in increasing the inference speed of generative AI to enable real-time applications. Its chatbot (based on the Mixtral or LLAMA2 LLMs) provides much faster response times than other chatbots on the queries I tested. I admit this may be a function of load, but it was noticeably faster. The company is most interesting because it is designing its own chips to support AI inference. And we all know from watching NVIDIA that this space is hot.