The Algorithmic Bridge
·
1 week ago
OpenAI's GPT-5.6 Sol scored 7.8% on the ARC-AGI-3 benchmark, a test designed to measure fluid intelligence through pattern recognition games that humans solve over 90% of the time. This represents a 20-fold improvement over GPT-5.5's 0.43% score three months earlier, and the model distinguishes itself by correctly identifying game mechanics before execution rather than simply executing learned patterns. The result suggests that further progress toward general intelligence requires improved reasoning scaffolding and planning rather than raw intelligence, since the model's failures occur in multi-step inference composition rather than perception.
ChinaTalk
·
1 week ago
Mythos-class AI models can identify software vulnerabilities so effectively that hidden backdoors installed by Chinese manufacturers would likely be discovered before exploitation, potentially undermining the national security rationale for bans on Chinese technology imports. The argument rests on the capability of these models to find flaws in any manufacturer's products regardless of origin, making it difficult for any nation to hide engineered access points. As a result, purchasing decisions could shift from being based on country of origin to factors like cost and performance, while Chinese companies would need to adopt transparent vulnerability disclosure practices to remain globally competitive.
Zvi (Don't Worry About the Vase)
·
1 week ago
OpenAI released GPT-5.6-Sol and upgraded its voice mode, while Anthropic extended Claude Fable 5 access through July 12, and Grok 4.5 launched with 1.5 trillion parameters at $2/$6 pricing. Early testers report Sol and Fable represent significant jumps over previous models, with users finding distinct advantages for each depending on task type. The new voice mode and model improvements are reshaping how people interact with AI systems and the competitive landscape among top model providers.
TheSequence
·
1 week ago
The article argues that verifiability alone is insufficient for determining whether a domain is suitable for AI development, proposing instead a multi-dimensional framework where domains like mathematics and chess excel because they score highly across multiple properties including grindability. The author contrasts high-performing domains such as code and board games with struggling domains like robotics and open-ended knowledge work, suggesting the latter fail on several unstated axes despite partial strength in others. This framework explains why AI systems have made faster progress in formal domains and why some reinforcement learning environment startups may ultimately disappoint investors despite their high valuations.
