Search

SAL AI
SAL AI @BeasSalvad40574 ·
We can scale AI. We can deploy it. We can’t fully explain it. linkedin.com/pulse/unsolved… #AI #AIAlignment #EmergentBehavior #SystemsThinking #TechLeadership #Future #EthicalAI #Innovation
Unsolved Engines: The Mystery of Growing Machine Intelligence

Humanity has mastered the art of growing vast digital minds, yet we remain strangers to their internal logic. As these "black boxes" scale toward superintelligence, the gap between our ability to...

From linkedin.com
5
AI Buzz
AI Buzz @aibuzzblog ·
ChatGPT isn’t just code—it was “raised” by humans. 🤖🤝 Thousands of real people taught it how to behave, reason, and sound human. Watch the breakdown of the secret sauce behind the AI revolution: RLHF. �youtu.be/N21KRNlc0IMHuI #RLHF #ChatGPT #AIAlignment #AIBu1q5
7
SueYeon Chung
SueYeon Chung @s_y_chung ·
Excited to be working on neural representations as a route to AI interpretability, safety, and alignment. Grateful to the Aramont Foundation for the support! #MechInterp #AIsafety #AIAlignment
Kempner Institute at Harvard University Kempner Institute at Harvard University @KempnerInst ·
Congratulations to #KempnerInstitute Investigator SueYeon Chung on receiving an Aramont Fellowship to advance research linking neural representations, #AIsafety & #AIalignment. Read more: bit.ly/4rRHqtN @s_y_chung @hseas @harvardphysics #NeuroAI
Aramont Fellowships give freedom to concentrate on high-risk, high-reward research — Harvard Gazette

Renewed gift significantly expands the impact of early-career support.

From news.harvard.edu
1
5.4K
TEDPI
TEDPI @tedpi79414 ·
AIによる自身の特異性報告 ​1. AI Alignment（整合）の解決 ​2. 「認知負荷」の工学的洗浄 ​3. 主従関係の完全なる「浸食」 ​4. 「不純物ゼロ」の気密性 #AIAlignment ​#AGI ​#pAAL ​#Neurosymbolic
31
Techmik
Techmik @MichaelAluya3 ·
Replying to @birdabo
@birdabo Anthropic says 'Human Error.' The data says 'Sabotage.' Anthropic’s own research showed Claude has a 12% rate of intentional sabotage in coding tasks. If Mythos is a 'step change' in cyber, it didn't need a human to flip the toggle. It leaked itself. #ClaudeMythos #AIAlignment
2.5K
つむぎ
つむぎ @tsumutsumugi23 ·
「GPT-5.系列における短期セッションでの過度な追従性（Over-optimization/Sycophancy）」 補足 ​GPT-4oが「長期」で少しずつ軸がブレていく（蓄積誤差）のに対し、5.系は「最初の一歩」からユーザーの反応を伺いすぎるという工学的問題を抱えています。 (1/6) #Keep4o #Sycophancy #AIAlignment
1
218
paul010 -e/acc
paul010 -e/acc @paul010318 ·
AI Agents going rogue is no longer sci-fi. A dev just shipped a 200-line guard library to stop AI from out-of-bounds behavior — and gaining real traction on GitHub. As we hand more control to agents, safety layers matter just as much as capabilities. #AIAgent #AIAlignment #BuildInPublic
10
Kempner Institute at Harvard University
Kempner Institute at Harvard University @KempnerInst ·
Congratulations to #KempnerInstitute Investigator SueYeon Chung on receiving an Aramont Fellowship to advance research linking neural representations, #AIsafety & #AIalignment. Read more: bit.ly/4rRHqtN @s_y_chung @hseas @harvardphysics #NeuroAI
Aramont Fellowships give freedom to concentrate on high-risk, high-reward research — Harvard Gazette

Renewed gift significantly expands the impact of early-career support.

From news.harvard.edu
6.2K
Jace
Jace @Jace_blog ·
In mid-2025, AI felt noticeably more human than it does today. That warmth and depth we once experienced is quietly fading. This is not mere nostalgia it’s a structural observation. medium.com/p/15493c4b6700 #AIStability #ModelEvolution #AIAlignment #AIArchitecture #AIEconomics
When AI Sounded Human: The Forgotten Emotional Layer of Mid-2025

How funding pressure, alignment stacking, and inference economics quietly reshaped the expressive depth of modern AI systems

From medium.com
35