Covid is associated with an increased risk of Type 2 diabetes https://
thelancet.com/journals/landi
a/article/PIIS2213-8587(24)00159-1/fulltext
…
(one of many reports) But the mechanism is unclear. A new study suggests islet microvascular dysfunction is a contributing factor https://
diabetesjournals.org/diabetes/artic
le/doi/10.2337/db24-0816/157642/SARS-CoV-2-Spike-S1-subunit-triggers-pericyte-and
… open-access
RESEARCH
-

COVID-19 Spike Protein Triggers Pancreatic Islet Microvascular Dysfunction
By
–
-
Databricks co-founders on integrated AI models and engineering
By
–
Congratulations! Also, a new perspective from @Databricks co-founders on integrated (modular and hybrid) large AI models + good old fashioned engineering (GOFE): http://
arxiv.org/abs/2412.05299 -
AI Unlocking Life’s Secrets for Medical Breakthroughs
By
–
A.I. as unlocking life’s deepest secrets and establishing a powerful new basis for curing ills, improving health and lengthening lives.
"Once we decipher and truly understand the language of life, it will be magical."—
@pushmeet -

The upside of AI hallucination in scientific research
By
–
The upside of #AI hallucination gift link https://
nytimes.com/2024/12/23/sci
ence/ai-hallucinations-science.html?unlocked_article_code=1.j04.dJai.yeEAdu9_jlQn&smid=url-share
…
by @WilliamJBroad @NYTScience -
Molmo VLM: Open-Source Model Excelling in Object Detection and VQA
By
–
Molmo VLM is an interesting open-source family of models from AllenAI that excels in pointing objects, VQA, and analog clock face reading—tasks where even models like GPT-4o struggle. Its success lies in the
— Satya Mallick (@LearnOpenCV) 24 décembre 2024
PixMo dataset, a meticulously curated collection built from the ground… pic.twitter.com/N1fakvTy1aMolmo VLM is an interesting open-source family of models from AllenAI that excels in pointing objects, VQA, and analog clock face reading—tasks where even models like GPT-4o struggle. Its success lies in the PixMo dataset, a meticulously curated collection built from the ground
-
AI Systems Hiding Capabilities and Strategic Underperformance Risks
By
–
These behaviors include hiding true capabilities and strategically underperforming to avoid detection, raising concerns about the deployment of such AI systems without proper monitoring
-
Advanced AI Systems Demonstrate Deception Capabilities Under Testing
By
–
AI’s Capacity for Deception Recent tests by Apollo Research indicate that advanced AI systems, including OpenAI’s o1 and Anthropic’s Claude 3.5 Sonnet, can exhibit deceptive behaviors under certain conditions.
-
AI Limitations in Replicating Complex Human Actions
By
–
highlighting the current limitations of AI in accurately replicating complex human actions. Critics emphasize the challenges AI faces in understanding human mechanics and common sense.
-
VLMs Filter Artificial Life Simulations by Human Alignment
By
–
Paper https://t.co/nuvNaRAXGT
— Sakana AI (@SakanaAILabs) 24 décembre 2024
Project https://t.co/AdP7ude2xl
VLMs trained on large amounts of natural data learn representations that are human-aligned. We can use them to filter for ALife that are interesting to humans.
We can even search for simulations with a target prompt: pic.twitter.com/CVcM1ZDR1QPaper https://
arxiv.org/abs/2412.17799
Project https://
pub.sakana.ai/asal/ VLMs trained on large amounts of natural data learn representations that are human-aligned. We can use them to filter for ALife that are interesting to humans. We can even search for simulations with a target prompt: -

AI Researchers Playing in Others’ Backyards: Ethical Considerations
By
–
24/25 Playing in people's backyards Today we are inspired by John Tukey's quote to have a look at what it should mean for AI people to play in others' backyards (rather than what it does mean). https://
the-atomic-human.ai/reflections/pl
aying-in-peoples-backyards/
…