Subscribe

Shane

Jan 19, 2026Jan 19

Category

Empty

1.

OpenAI was reported to have solved another Erdős problem with GPT-5.2 Pro, while a new database revealed that most AI attempts at comparable mathematical problems had an estimated one-to-two percent success rate.

2.

Snap's SnapGen++ generated high-resolution AI images on iPhone in under two seconds, running server-quality image generation directly on phones and outperforming models roughly 30 times larger despite using about 0.4 billion parameters.

3.

Google upgraded AI Overviews to route complex search queries to Gemini 3 Pro while simpler queries continued to be handled by faster models, and the change was limited to paying subscribers.

4.

Researchers found that even the most capable multimodal language models failed at basic visual tasks that toddlers mastered before they learned to speak.

5.

South Koreans were reported to have spent more on AI subscriptions than on Netflix each month, with ChatGPT dominating subscription spending.

References

1.

https://the-decoder.com/gpt-5-2-pro-solves-another-erdos-problem-while-a-new-database-reveals-most-attempts-still-fail/

GPT-5.2 Pro solves another Erdős problem while a new database reveals most attempts still fail

GPT-5.2 Pro has likely solved another Erdős problem. But mathematician Terence Tao warns against a skewed perception: AI's actual success rate on such problems sits at just one to two percent.

the-decoder.com

1.

https://the-decoder.com/snaps-snapgen-generates-high-resolution-ai-images-on-iphone-in-under-two-seconds/

Snap's SnapGen++ generates high-resolution AI images on iPhone in under two seconds

Snap's SnapGen++ runs server-quality image generation directly on phones. Despite having just 0.4 billion parameters, it beats models 30 times larger.

the-decoder.com

1.

https://the-decoder.com/google-upgrades-ai-overviews-with-gemini-3-pro-for-complex-queries/

Google upgrades AI Overviews with Gemini 3 Pro for complex queries

Google now routes complex search queries in AI Overviews to its most capable model, Gemini 3 Pro, while simpler questions still go to faster models. For now, the feature is limited to paying subscribers.

the-decoder.com

1.

https://the-decoder.com/even-the-best-ai-models-fail-at-visual-tasks-toddlers-handle-easily/

Even the best AI models fail at visual tasks toddlers handle easily

A new study exposes a fundamental weakness in today's AI systems: even the most capable multimodal language models can't handle basic visual tasks that toddlers master before they learn to speak.

the-decoder.com

1.

https://the-decoder.com/south-koreans-now-spend-more-on-ai-subscriptions-than-netflix-each-month/

South Koreans now spend more on AI subscriptions than Netflix each month

Monthly AI subscription spending in South Korea has surpassed Netflix. ChatGPT dominates.

the-decoder.com

more pain more gain 🚀

© 2024-2025 Shane "Lx". All rights reserved.