Skip to content
Longterm Wiki
Back

Conjecture - AI Safety Research Blog

web
conjecture.dev·conjecture.dev/

Conjecture is a UK-based AI safety company pursuing the cognitive emulation research agenda; their blog is a primary source for understanding CoEm and related technical safety work.

Metadata

Importance: 45/100blog posthomepage

Summary

Conjecture is an AI safety research company focused on cognitive emulation (CoEm) as an approach to building aligned AI systems. Their blog covers technical AI safety research, interpretability, and alignment strategies with a particular emphasis on making AI systems that reason more like humans in interpretable ways.

Key Points

  • Conjecture develops Cognitive Emulation (CoEm), an approach to alignment that aims to build AI systems mimicking human cognitive processes
  • Research focus includes interpretability, understanding AI internals, and developing safer training paradigms
  • The company takes a commercial approach to AI safety, attempting to make safety-focused AI economically viable
  • Blog covers both technical research and broader strategic thinking about AI risk and alignment
  • Conjecture was founded by former EleutherAI and other AI research community members

Cited by 2 pages

PageTypeQuality
ConjectureOrganization37.0
Survival and Flourishing Fund (SFF)Organization59.0

5 FactBase facts citing this source

EntityPropertyValueAs Of
ConjectureFounded DateMar 2022
ConjectureTotal Funding Raised$25MDec 2022
ConjectureFounded Bysid_CrXoCsIucX, sid_kLIpOZtU1n, sid_n0d7I3OAej
ConjectureHeadquartersLondon, UK
ConjectureLegal StructurePrivate company

Cached Content Preview

HTTP 200Fetched Apr 13, 20265 KB
Conjecture 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 Product 

 Research 

 About Us 

 Contact 

 Product 

 About Us 

 Contact 

 Research 

 Product 

 Research 

 About Us 

 Contact 

 Product 

 Research 

 About Us 

 Contact 

 Home 

 Redefining AI Safety

 Redefining AI Safety

 Redefining AI Safety

 Building a new AI architecture to ensure the controllable,

 safe development of advanced AI technology.

 Building a new AI architecture to ensure the controllable, safe development of advanced AI technology.

 Building a new AI architecture to ensure the controllable, safe development of advanced AI technology.

 Request a Demo

 Learn More

 Navigating Complexities

 The challenge of AI Safety

 The challenge of AI Safety

 The challenge of AI Safety

 Learn More

 Unpredictable

 AI systems generate hallucinations and inadvertently leak sensitive information, compromising their reliability.

 Incoherent

 AI responses are inconsistent in their outputs and reasoning, creating challenges to effective interaction.

 Inept

 Systems fail on basic tasks, raising significant obstacles to building reliable automation.

 Uninterpretable

 AI's inner workings remain uninterpretable, making it difficult to trust the accuracy of its outputs and debug when it’s incorrect.

 Revolutionizing AI Deployment

 Revolutionizing AI Deployment

 Introducing Cognitive Emulation

 Introducing Cognitive Emulation

 Introducing Cognitive Emulation

 Building and deploying AI systems that are both powerful and safe faces great challenges in the current AI paradigm. And so, we are building Cognitive Emulation: A different vision for powerful AI systems that are designed to follow the same, trusted reasoning processes we do.

 Learn More

 Learn More

 Learn More

 Learn More

 Trained for specific tasks

 Trained for specific tasks

 Build AI by component

 Build AI by component

 Automate real workflows

 Automate real workflows

 Solve more complex problems

 Solve more complex problems

 Capability vs. Safety

 Scaling and the Control Problem

 Scaling and the Control Problem

 The AI industry is racing to scale ever-larger models without considering the risks. While capabilities advance at a rapid pace, safety lags far behind. The current imbalance underscores the urgency for innovative solutions like Cognitive Emulation as an alternative to the grave risks associated with scaling.

 The AI industry is racing to scale ever-larger models without considering the risks

 Amplified Risks

 Amplified Risks

 Scaling Exacerbates the Dangers

 Scaling Exacerbates the Dangers

 Scaling Exacerbates the Dangers

 Amplifying these models through scaling only makes it harder to notice if they are wrong, and impossible to debug when they are. 

 Amplifying these models through scaling only makes it harder to notice if they are wrong, and impossible to debug when they are. 

 Amplifying these models through scaling only make

... (truncated, 5 KB total)
Resource ID: b7aa1f2c839b5ee8 | Stable ID: sid_2aCx13LY7g