Happy Wednesday, Lattice Community!
We’ll dissect big stories in AI this week, provide insights to keep you informed in your industry, and share updates on our work. Lattice Partners is an AI consulting firm that helps businesses with AI strategy and custom implementations.
This week’s TL;DR:
• A researcher’s resignation letter about AI developing too fast and what this means for the future of humanity
• Mapping a fly’s brain using AI could help us understand diseases like Alzheimer’s
• An AI model solved one of the hardest, unresolved math problems in the world
• What personal data is safe for me to share with my AI tool?
An ex-researcher at OpenAI and Anthropic is claiming that AI could kill all humans by 2030.
Jacob Coxon spent three years building AI models at OpenAI and Anthropic. In early September, he quit his job and posted a public warning that leading AI labs are developing too fast and will outthink and outsmart humans soon.
His post went viral in 24 hours. One of OpenAI’s top scientists echoed a similar sentiment, calling on the industry to slow down until there are more guardrails in place for this technology.
Takeaways
Lawmakers are responding quickly
One new bill requires safety testing and reporting to happen prior to new AI model releases. Another bill proposes pausing development of the most powerful AI systems until the government enacts safety measures and legislation.
Geopolitical consequences
If American companies slow down development while other countries continue at full speed, the next challenge is getting the rest of the world to agree on the limits together.
Let’s take a step back
It’s easy to read headlines and panic, but 50 years ago, today’s weapons also would have looked like they could guarantee mass destruction. The world built ways to defend against them - we think AI is likely to follow a similar trend. We need to build it carefully, and establish safety nets simultaneously while we develop.
Can AI actually give us hope for curing diseases like Alzheimer’s?
Scientists mapped an entire fly’s brain.
Researchers at Google, in collaboration with the science institute HHMI Janelia, mapped every neuron in a male fruit fly’s brain (166,000) and every connection between them (125 million). This is the largest brain map to date. Scientists have worked to map the adult fruit fly brain since 2008, meaning it’s taken nearly 18 years of intense work to reach the complete central nervous system map released last week.
Takeaways
AI made this possible
Mapping 125 million connections by hand is not realistic. Research teams used a process called connectomics, or brain mapping. It works by sectioning a brain into thin layers, imaging each one, then using AI to connect the images into a 3D map of each neuron and connection.
Medical research impacts
Although a human brain has around 86 billion neurons, researchers can utilize a similar approach to better understand diseases like Alzheimer’s and establish the groundwork for treating them.
We dive into all these topics more on the Lattice Partners podcast, the AI Playbook, linked here.
AI solved a million-dollar math problem that’s stumped researchers for 90 years.
Navier-Stokes is a famous math problem that focuses on how liquids and gases move, which is important for airplane design and weather forecasting.
There are seven “Millennium Prize” problems in the world, each with a million-dollar prize as the reward for solving it. Only one had ever been solved before. OpenAI utilized thousands of AI systems, working simultaneously, along with an estimated $6 million in credits to solve Navier-Stokes in under four days. This announcement has since been scrutinized over a credit controversy.
Takeaways
Science advances due to the process, not just the solution
Discoveries made along the way when solving complex problems are just as valuable as the solve itself. Terence Tao, a Fields Medal winner at UCLA and widely known as the “best living mathematician,” warns that AI companies rushing to utilize unpublished progress could cause scientists to be more secretive about their early work. This would be a negative, backwards outcome.
Nothing is official. The Clay Mathematics Institute says its process to review the proof will be “deliberately unhurried.”
A new benchmark
GPT-6 Astra was put to a sillier use inside Blender, a 3D animation tool, by attempting to recreate the old “Will Smith eating spaghetti” AI video that went viral for how bad it looked. People are using this as an informal benchmark to track how fast these AI models are improving.
Read more here: OpenAI’s post and Axios on the result.
Something you can use
Is Claude keeping my data?
• Check the platform’s data policy and opt out of training if the option exists.
• You can treat it like your email or text messages. You share plenty of personal information there and trust the company holding your data to keep it safe. The same thing applies here.
• Keep Social Security numbers, credit card numbers, or anything that could personally identify you out of the chat.
This week at Lattice Partners
We’re building MCPs for many of our clients!
Lattice Partners is an AI consulting firm that helps businesses leverage the power of AI through strategic consulting, hands-on training, and custom implementations.
We’ve been building MCPs (Model Context Protocol). You can think of these as translators that allow Claude or ChatGPT to talk directly to other parts of your business. We’re creating these for older systems that many companies are still running their business on. Acumatica is one example - an ERP that handles accounting, inventory, and operations, but was never built as an AI-native system.
Once we build an MCP for a client, they’re able to ask Claude or ChatGPT to pull reports, check orders or update records directly from the chat. The MCP talks to a system like Acumatica. Our clients never have to dig through files and folders again.
If your team could benefit from something like this, feel free to reach out to us and we’d be happy to help.
See you next Wednesday!
Cheers,
Lattice Partners
