MIT CAMBRIDGE, MASSACHUSETTS
Getty Images
There’s so much going on, it’s sometimes hard to keep track of it all. You can get some guidance, though, from surveys of the biggest emerging projects, presented by people close to the industry. I was pleased to see MIT’s list of notable tech advances for the year ahead, and wanted to mention these according to their applications to our world.
Whether they’re related to energy, model discovery or hardware innovation, these are things you’re probably going to hear a lot more about as the new year commences.
Greener Batteries, Safe Nuclear and Energy Advances
A few of the items on the list here at the MIT slate of “10 Breakthrough Technologies” have to do with meeting the energy challenges that we face as societies. Our demand for power is vast, not least because of the enormous appetite of LLMs served by massive data centers. To that end, companies like Constellation, Vistra, TerraPower and others are pioneering new forms of nuclear energy to co-locate with the data centers, in order to minimize transfer-related loss.
And solar energy is ubiquitous, but how do you store it? Scientists are working on sodium-ion batteries that solve several problems with lithium-ion batteries currently used in devices, including cost, and the nature of lithium mining.
Generative Coding and Agentic Task-Handling
This one is so evident it almost goes without saying: over the course of 2025, we’ve seen LLMs go from relatively simple answer-machines, to juggernauts that can manipulate tools, navigate systems, and accomplish tasks like people.
“The barrier to entry isn’t learning syntax anymore,” writes an analyst at Orbit, specifically citing the MIT resource. “It’s learning to describe what you want clearly.”
So only months after Andrej Karpathy notably coined the term “vibe-coding,” here we are. And you can expect all of this to continue in 2026.
Breakthroughs in Genetics
This is another category that got more than one mention on MIT’s list, which included genetic work on “resurrecting” extinct species (I wrote about this as applied to direwolves early last year), and a gene-edited baby named… KJ?
A press release from Children’s Hospital of Philadelphia goes into detail about how scientists Rebecca Ahrens-Nicklas and Kiran Musunuru worked on this solution.
“After years of preclinical research with similar disease-causing variants, Ahrens-Nicklas and Musunuru targeted KJ’s specific variant of CPS1, identified soon after his birth,” write spokespersons. “Within six months, their team designed and manufactured a base editing therapy delivered via lipid nanoparticles to the liver in order to correct KJ’s faulty enzyme. In late February 2025, KJ received his first infusion of this experimental therapy, and since then, he has received follow-up doses in March and April 2025.”
That’s exciting research, and it could pave the way for a lot of similar interventions in the future, “cures” based on genetic changes closing the doors on pathologies that were, in a sense, encoded into the DNA.
Working Robots and Companion Robots
Another slot on the list went to “companion robots,” the flip side of a pretty advanced set of applications for laboring robots in industry. But here’s how the MIT Technology Review treated this entry:
“Every day, millions of people interact with AI chatbots. Some of them form what feel like close, personal bonds with the bots. There’s mounting evidence that this can be dangerous, and politicians are finally waking up.”
Many reports of concern revolve around a number of teen suicides potentially linked to LLM use, but there are other possible effects as well.
The state of California, for one, is not waiting around.
“California has enacted the first U.S. law mandating safety standards for AI companion chatbots after several suicide cases involving young users, including incidents linked to OpenAI, Meta, and Character AI,” writes Maximilian Schreiner at The Decoder. “The law, SB 243, takes effect on January 1, 2026, and requires platforms to verify users’ ages, display warnings, disclose that conversations are artificial, block sexual content for minors, and implement crisis response protocols shared with the state health department.”
Keep an eye on this.
The Promise of Mechanistic Interpretability
In 2026, computer scientists will also be working on the sizable job of mapping out LLM responses to try to figure out more about how our model friends “think.”
Here’s a resource from Anthropic, again, not specifically authored, that goes into detail on how this works, using examples like the San Francisco Bay Bridge and revealing token relationships and resulting neural attention patterns.
“Mechanistic interpretability seeks to reverse engineer neural networks, similar to how one might reverse engineer a compiled binary computer program,” writes Chris Olah in a “note” that goes over a lot of the involved methodology.
Or take this explainer metaphor from Demba Ba at the Kempner Institute:
“Imagine yourself in a Boston Symphony Orchestra performance, closing your eyes, and trying to guess which set of instruments are playing in any particular snippet of time: some might be always present in the background, while others might have sparser contributions to the musical piece. This might be an accessible task to a highly trained musician, but a nearly impossible task for someone without a musical background. This is a point source detection problem, in which the multiple sources of a complex signal are located. The brain is somewhat like an orchestra: a large ensemble of cells of different types whose coordinated yet heterogeneous activity patterns generate (mostly) harmonious behavior.”
Essentially, this new method of analysis lets us get deeper under the hood, and understand neural nets in important ways. All of this promises to revolutionize our research on the LLMs and models that drive high-end applications and disrupt our world.
But how will all of this work? Will it be an orderly process, or something more akin to a technological Guernica? Stay tuned.
link