or
New to LinkedIn? Join now
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Diligent Robotics
Navera
VMS
Motorola Solutions
FinalLayer
Simbian
Y Combinator
Securin Inc.
Lumia Security
Intezer Labs
JusticeText
miniOrange
XTX Markets
Nucleus Security
Stony Brook University
Whisker
Robby Technologies
Lithic
Solana Foundation
American Express
Saif Islam
Savant Analytics • 3K followers
Ruha Benjamin on NYC's Technology Committee: "Technology is never neutral, and its harms and benefits are unevenly distributed." This is the foundation for progressive AI governance. Transparency isn't just about disclosing algorithms - it's about who controls the infrastructure, who audits the outcomes, and whether communities can challenge automated decisions that affect their lives. Progressive AI governance puts community power before corporate convenience. NYC can demonstrate what that looks like operationally. https://lnkd.in/ePaCibHy Ruha Benjamin Maia Woluchem Amba Kak Sam Jacobs Sarah Aoun Myaisha Hayes #AIGovernance #TechPolicy #NYC #AlgorithmicAccountability
Darrell Long
Institute for Defense Analyses • 3K followers
Benchmarks are not an indication of thinking. I decided to run some local models, and to my complete lack of surprise, Deep Seek has Chinese values deeply baked in, just like ChatGPT 5.2 has arrogant pedantic puritan deeply baked in. These LLMs are useful stochastic parrots; they are not emergent intelligence.
Guy Lebanon
Meta • 24K followers
Cool recent paper from our team (Instagram Ads Ranking) on customizing the learning process of attention layers to preserve mutual information with minority cohorts while improving global performance. This could be quite relevant to other industry groups working on large scale recommendation systems. https://lnkd.in/gyfVWbS5
Rhonda Coleman Albazie
PRIVILEGE HEALTH ™️ -… • 487 followers
What it implies for AI (now): Keep the American innovation ecosystem but internalize the learning loop, evaluation authority, and core technical cadre (U.S. Military AI Corps), so the country isn’t “renting cognition.”
Jason Upchurch, PhD
Dr. Jason Upchurch is an… • 925 followers
In July 2024, our team at RTI wrote an internal paper proposing "Software 3.0" - a thesis extending Andrej Karpathy's Software 2.0 concept by taking what AI systems learn from experience, persisting that knowledge as distributed services, and letting other agents reuse it without human intervention. At the time, it was just a thesis. No framework. No data. Two years later, we have both. In a beautiful synergy we took Karpathy's own AutoResearch work and realized Software 3.0. The Genesis AutoResearch experiment ran 75 fully autonomous iterations of a coding benchmark: hard DDS programming tasks that frontier AI models genuinely cannot solve out of the box. A teacher AI agent struggled with these tasks, discovered solutions, encoded them as distributed services on our Genesis framework, and from that point forward, every student agent solved those same tasks reliably. The results: • Claude Opus 4.6 went from a 33% baseline pass rate to 91% with Genesis tools, including 32 consecutive perfect runs • Claude Haiku 4.5 (the smallest, cheapest model) went from 0% to 97% using the exact same tools with zero modifications • Cost per task dropped from $0.17 to $0.007 - a 96% reduction That last point is worth sitting with. A model that costs 25x less, paired with the right infrastructure, outperformed a model 25x more expensive without it. The value isn't in the model. It's in the infrastructure around it. This has real implications for how organizations should invest in AI capability. Rather than always reaching for the largest, most expensive model, building domain-specific tool infrastructure and deploying it through a framework like Genesis may yield better results at dramatically lower cost. The three pillars of our Software 3.0 vision: lifelong learning, data-centric architecture, and real autonomy; have each been validated with quantitative evidence. Knowledge discovered by one agent, encoded as a distributed service, made all future agents more capable regardless of model size or cost. Full paper and experiment data: https://lnkd.in/gN5DjZwf Genesis framework: https://lnkd.in/gKiFagXZ
Find curated posts and insights for relevant topics all in one place.