About Us

At Conjecture, we believe that transformative Artificial Intelligence will be developed within the next decade. There is currently no solution to make AI safe.

Our History

Conjecture was founded in March 2022 by Connor Leahy, Sid Black, and Gabriel Alfour. Conjecture was born out of EleutherAI, also co-founded by Connor and Sid in 2020. One of the earliest and most successful open-source LLM communities, EleutherAI was responsible for developing, among many other things, GPT-J and GPT-NeoX, the largest OS LLM releases at the time.

EleutherAI was also a home for early discussions on the risks of LLM-based advanced AI systems. We have always believed that the core problem of advanced AI is the control problem, and this is what we set out to solve when founding Conjecture.

Today, Conjecture has a dedicated team of researchers and engineers working on current state of the art AI, building both our suite of state of the art LLMs and our vision of Cognitive Emulation atop this foundation. We are building the infrastructure that allows us to build scalable, auditable, controllable AI systems, and work to enable our customers to benefit from this new paradigm.

Our Vision

Few stories of great invention are better known than the story of the Wright brothers’ quest for flight. At that time, the Wright brothers were competing to build aircraft with a number of other inventors, but they had one distinct advantage that put them ahead.

What the Wright brothers understood was that the true difficulty in getting to self powered human flight was the danger to the pilot during testing, or in other words, the control problem.

In a letter, Wilbur Wright wrote:

“When once a machine is under proper control under all conditions, the motor problem will be quickly solved. A failure of a motor will then mean simply a slow descent and safe landing instead of a disastrous fall.”

Wilbur’s insight was that after designing a safe glider that will fail gracefully and non-lethally, in all conditions, the problem of building a lightweight and powerful enough engine is almost a trivial afterthought. The actual problem to be solved is one of safety and control.

History today remembers the Wright brothers because this approach made them win, and laid the groundwork for us all to enjoy extraordinarily safe and reliable air travel. While other inventors neglected the fundamental questions of airborne control and non-lethal crash landing (many brilliant competitors dying tragically while testing their own creations as a result), the Wright brothers’ gliders created a safe architecture around the engine, and enabled iterative improvements.

AI safety needs to be approached in the same way. LLMs are powerful engines, but they are just that, not a “safe glider.” An engine alone may be able to achieve takeoff, but it is not a reliable way of ensuring a safe landing or travel to one’s destination.

At Conjecture, we are in the business of building “airplanes”, not “engines.” (Though we make the necessary LLM “engines” as well!) We build systems that give us the control needed to enjoy the benefits of powerful LLMs while avoiding the potentially catastrophic downsides.

At Conjecture, our Cognitive Emulation approach is our roadmap towards exactly such an AI “airplane”. You can read more about our plan on our Research page. This approach lines up well with demands from users and enterprises who continue to look for models that act in predictable ways, a problem we solve with CoEm systems.

Are you as excited as we are to build the future of AI takeoff and landing? Come join us.

Come work with us!

Check out our current open positions!