The hottest Research Substack posts right now

And their main takeaways
Category
Top Health Politics Topics

Import AI 336: Financialized AI; public and elite AI opinion; one million insects.

Import AI • 519 implied HN points • 14 Aug 23
🕹 Technology Research
  1. The financialization of AI is increasing, with companies finding new ways to fund AI projects through unconventional means like debt collateralized against GPUs.
  2. AI benchmarks are being solved faster, indicating either accelerated AI progress or the increasing complexity in building good benchmarks.
  3. Public opinion, reflected in a poll, shows significant concerns about AI development and regulation, contrasting with elite opinions that emphasize rapid AI advancement.

Against the Burden of Knowledge

Maximum Progress • 294 implied HN points • 17 Jan 24
🔬 Science Research
  1. The burden of knowledge suggests that finding new ideas gets harder because you have to spend more time learning old ones.
  2. Knowledge is not always cumulative in scientific progress, as seen in historical examples like the transition from Ptolemy's geocentric model to Copernicus' heliocentric model.
  3. Institutional decay in academia may be a more plausible explanation for aging researchers, larger teams, and narrower fields, rather than just the burden of knowledge.

SHAP Is Not All You Need

Mindful Modeler • 898 implied HN points • 07 Feb 23
🕹 Technology Research
  1. It's important to avoid assuming one method is always the best for all interpretation contexts when working with machine learning interpretability tools like SHAP.
  2. Different interpretability methods like SHAP and permutation feature importance (PFI) have unique goals and can provide different insights, so it's crucial to choose the method that aligns with the specific question you want to answer.
  3. Research on interpretability should be more driven by questions rather than methods, to ensure that the tools used provide meaningful insights based on the context.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

Import AI 346: Human-like meta-learning; a 3 trillion token dataset; spies VS AI

Import AI • 459 implied HN points • 30 Oct 23
🕹 Technology Research
  1. UK's intelligence services are slightly worried about the safety implications of generative AI technologies, particularly in amplifying existing risks like cyber-attacks and digital vulnerabilities
  2. Research shows that a basic Transformer neural net architecture can meta-learn and match human performance in inferring complex rules from small data, hinting at AI systems increasingly displaying human-like qualities
  3. Facebook's Habitat 3.0 software enables training and testing agents to collaborate with humans by simulating realistic 3D environments with humanoid avatars, human-in-the-loop interactions, and benchmark tasks for human-robot interaction

A Massive Win, and What It Means

The Crucial Years • 2002 implied HN points • 26 Jan 24
🌞 Climate & Environment Research
  1. A huge victory against the fossil fuel industry has been achieved by halting new licenses for LNG export terminals.
  2. Biden's decision sets a climate test for American fossil fuel expansion plans, signaling a transition from natural gas as a 'bridge fuel'.
  3. The fight for climate justice needs to include considerations for impacts on communities and involve effective monitoring and engagement with frontline leaders.

The Harm of Weight Stigma

Weight and Healthcare • 678 implied HN points • 08 Mar 23
🏥 Health & Wellness Research
  1. Weight stigma can have significant negative impacts on physical and mental health, including increased risk for conditions like high blood pressure, diabetes, and poor health behaviors.
  2. Perceived weight discrimination is linked to psychiatric disorders, risk of suicidality, and even an increase in mortality risk, demonstrating the serious harm caused by weight stigma.
  3. Healthcare practitioners, including doctors and nurses, may exhibit weight bias which can lead to barriers in accessing healthcare and biased treatment, affecting the well-being of higher-weight individuals.

From Intentions to Actions: Achieving Goals with Implementation Intentions

UX Psychology • 297 implied HN points • 05 Jan 24
🏥 Health & Wellness Research
  1. Form implementation intentions to bridge the gap between intentions and actions by creating specific if-then plans linking cues to goal-directed behaviors.
  2. Research shows that forming if-then plans can significantly increase the likelihood of achieving goals in various areas such as health, consumer choices, and personal goals.
  3. To apply implementation intentions, choose specific goals, identify situational cues, specify goal-directed responses, form if-then plan statements, and commit to following through.

The Truth About ObesityWeek - Part 2

Weight and Healthcare • 439 implied HN points • 28 Oct 23
🏥 Health & Wellness Research
  1. ObsityWeek focuses on promoting the weight-centric paradigm in healthcare by connecting weight with health and disease, potentially leading to harmful practices.
  2. Corporate influence is prevalent at ObsityWeek through sponsored symposia and key lectures, where pharmaceutical companies push for weight-centric treatments like medications and surgeries.
  3. There is a disconnect at ObsityWeek between celebrating weight loss industry figures and the lack of progress in reducing the population of higher-weight individuals, highlighting the profit-driven nature of the field.

Two cautionary tales about BA.2.86

Logging the World • 478 implied HN points • 07 Oct 23
🔬 Science Research
  1. Don't just look at proportions, consider sample sizes too. Confidence intervals are important when dealing with limited data.
  2. Focus on the bigger picture when it comes to analyzing new variants. Growth rates matter.
  3. Instead of fixating on one variant, keep an eye on other potentially impactful variants as well. The situation is constantly evolving.

Never mind the quality, feel the speed

imperfect offerings • 239 implied HN points • 02 Feb 24
🚌 Education Research
  1. The research economy is increasingly focused on speed over quality, especially with the rise of generative AI, which can have negative impacts on reproducibility and diverse fields of knowledge.
  2. Data models in research need to be carefully scrutinized for accuracy and not blindly relied upon, even in specialized areas like protein folding, climate science, or medical diagnostics.
  3. Speed and heuristics shouldn't overshadow the importance of deliberation, qualitative research, and embracing complexity in arriving at meaningful solutions to multidimensional problems.

Genetic Variations that Predispose People to Severe COVID-19, Cause of Acute Deadly Hepatitis in Children Likely Identified, SARS-CoV-2 Origin Still Being Investigated

LIL Science • 471 implied HN points • 18 May 23
🔬 Science Research
  1. Research has identified genetic variations that make people more likely to have severe COVID-19.
  2. The likely cause of acute deadly hepatitis in children has been identified as a combination of viruses and genetic predispositions.
  3. The origin of SARS-CoV-2 is still under investigation, with reports detailing activities in China preceding the pandemic.

What’s happening inside the brain on LSD? Colorado psychedelics legislation flies through state Senate, and Oregon gets its first licensed testing lab

The Microdose • 471 implied HN points • 28 Apr 23
🏥 Health & Wellness Research
  1. Scientists study brain activity on LSD and find altered connectivity and self-inhibition.
  2. Colorado Senate passes SB 290 to implement Proposition 122 for psilocybin services in the state.
  3. Oregon licenses its first psilocybin testing lab, but high costs may drive some practitioners underground.

Do red and processed meats cause cancer?

Unsettled Science • 840 implied HN points • 21 Oct 24
🏥 Health Politics Research
  1. The World Health Organization says processed meats cause cancer, but only one type of cancer has been linked directly to them, which is colorectal cancer.
  2. The IARC's decision was based on a small number of studies, which raises questions about how strong the evidence really is.
  3. Many people reacted strongly to the news, fearing that all red meats are dangerous, but the evidence used may not support that broad conclusion.

Losing my Religion, 2e

KCKlatt’s Substack • 2 HN points • 22 Sep 24
🚌 Education Research
  1. Navigating academic life can create a lot of confusion, especially when you see a gap between the ideals of research and the reality of how things operate. It can feel frustrating when the focus is more on getting published than on genuine discoveries.
  2. The pressure to publish often leads to questionable practices in research, which can hurt the integrity of the work. Many people in academia are just trying to survive in a system that doesn’t set them up for real success.
  3. Finding a supportive environment in academia is tough, and it's important to constantly question what the true mission of academic work is. Many academics feel like they’re in a game where they have to compete rather than collaborate.

Altered states: The new science of appetite and brain plasticity

Brain Pizza • 662 implied HN points • 23 Dec 24
🏥 Health & Wellness Research
  1. New weight-loss drugs like semaglutide can help people lose weight and improve their overall health. They might change how we think about appetite and cravings.
  2. These medications could have benefits beyond just weight loss, including helping with anxiety and even substance use disorders. They show promise in many areas of health.
  3. There is potential for these drugs to change society by reducing healthcare costs, improving self-esteem, and even aiding in the treatment of diseases like Parkinson’s.

Epistemic Hell

Secretum Secretorum • 1768 implied HN points • 27 Jan 24
🔬 Science Research
  1. Through history, reasoning based on limited information has led to dangerous outcomes, like chronic cyanide poisoning and scurvy.
  2. Some scientific problems, like understanding heredity, have been challenging due to complexity and the need for exceptional datasets.
  3. To escape epistemic hells, radical ideas, serendipity, unlearning prior knowledge, and persistence are vital.

The Sequence Radar #779: The Inference Wars and China’s AI IPO Race

TheSequence • 35 implied HN points • 28 Dec 25
🕹 Technology Research
  1. Nvidia licensed Groq’s LPU technology and brought key Groq leaders onboard, consolidating talent and inference IP to reinforce its lead in inference hardware.
  2. Chinese model labs are shipping frontier models: Zhipu’s GLM 4.7 pushes coding and agentic ‘deep thinking,’ while MiniMax’s M2.1 uses linear attention and MoE to enable a massive 4‑million‑token context window at much lower cost.
  3. Zhipu and MiniMax preparing Hong Kong IPOs shows foundation models are moving from VC-funded research to public, revenue-focused companies, and highlights a split where U.S. scaling is driven by capital and hardware consolidation while China focuses on architectural and economic efficiency.

Benzos (benzodiazepines) Are Associated With Death By Suicide.

The Frontier Psychiatrists • 255 implied HN points • 12 Jan 24
🏥 Health & Wellness Research
  1. Benzodiazepines like Ativan and Xanax are linked to higher risk of suicide, even in individuals without prior psychiatric history.
  2. Studies have consistently shown an increased risk of suicide with benzodiazepine exposure.
  3. Alternative treatments to benzodiazepines are needed for anxiety to prevent the heightened suicide risk associated with these medications.

THB Insider #12

The Honest Broker Newsletter • 628 implied HN points • 19 Dec 24
🔬 Science Research
  1. The author will soon evaluate their predictions for 2024 and will post about the upcoming 2025 predictions. It's a fun way to look back and see how accurate they were.
  2. A new paper has been posted for feedback and will be revised based on comments received. This is a chance for readers to engage and contribute their thoughts.
  3. The author is planning a Q&A session and is inviting readers to submit their questions. It's an opportunity for direct interaction and to discuss important topics.

Some News

Cybernetic Forests • 159 implied HN points • 12 Mar 24
🎨 Art & Illustration Research
  1. Eryk Salvaggio has been named a 2024 Research Fellow with the Flickr Foundation, an organization preserving shared visual content for the future.
  2. Their research project will focus on AI-generated images and exploring Flickr's archives for training data.
  3. Eryk Salvaggio will be in London for a one-month residency in April 2024, looking forward to meeting people and sharing insights on their work.

Evo 2 Can Design Entire Genomes

Asimov Press • 490 implied HN points • 19 Feb 25
🕹 Technology Research
  1. Evo 2 is a powerful AI model that can design entire genomes and predict harmful genetic mutations quickly. It can help scientists understand genetics better and improve genetic engineering.
  2. Unlike earlier models, Evo 2 can analyze large genetic sequences and understand their relationships, making it easier to see how genes interact in living organisms.
  3. While Evo 2 offers exciting possibilities for bioengineering, there are also concerns about its potential misuse. It's important to handle such powerful technology responsibly to avoid harmful applications.

BA.2.86 growth rates

Logging the World • 418 implied HN points • 23 Aug 23
🔬 Science Research
  1. New COVID variant BA.2.86 has mutations that suggest fast growth, but estimating its growth rate is tricky.
  2. Statisticians use models and likelihood functions to estimate parameters like growth rates, but uncertainty exists in the estimates.
  3. The work of statistician C.R. Rao, like the Fisher information, shows fundamental limits to parameter estimation and the role of geometry in statistics.

The Sequence Radar #477: The R1 Moment

TheSequence • 546 implied HN points • 26 Jan 25
🕹 Technology Research
  1. DeepSeek-R1 is a new AI model that shows it can perform as well or better than big-name AI models but at a much lower cost. This means smaller companies can now compete in AI innovation without needing huge budgets.
  2. The way DeepSeek-R1 is trained is different from traditional methods. It uses a new approach called reinforcement learning, which helps the model learn smarter reasoning skills without needing a ton of supervised data.
  3. The open-source nature of DeepSeek-R1 means anyone can access and use the code for free. This encourages collaboration and allows more people to innovate in AI, making technology more accessible to everyone.

What Humans Do Next

Working Theorys • 242 implied HN points • 25 Jun 25
🕹 Technology Research
  1. In the future, jobs will shift from traditional roles to five main areas: trades, research, art, community, and stewardship. We will still work, but the nature of our work will change.
  2. Art won't just be about creating traditional pieces; it will involve everything that brings beauty and connection into our lives. The human touch in creativity will remain important, and art will become more personal and community-focused.
  3. Community building will become a vital job as more people seek local connections. It's about creating spaces where people can come together, and this type of work will be valued more than many tech-driven roles.

"Everything that moves in the future will be robotic."

Robots & Startups • 139 implied HN points • 24 Mar 24
🕹 Technology Research
  1. The future is predicted to be dominated by robotic technology, with everything that moves expected to be robotic.
  2. NVIDIA CEO Jensen Huang envisions the Omniverse as the future operating system and aims to make computers smarter for easier human interactions.
  3. The robotics industry is rapidly advancing, with developments in humanoid robots, AI, and innovative collaborations between humans and robots promising significant changes in various sectors like labor, healthcare, and space exploration.

The Validation and Frustration of Stunkard et al.

Weight and Healthcare • 619 implied HN points • 11 Feb 23
🏥 Health & Wellness Research
  1. Stunkard et al.'s 1959 study on weight loss interventions revealed the ineffectiveness and harmfulness of most programs from as early as the 1950s.
  2. The study highlighted common flaws in weight loss research, such as short-duration studies, lack of specific data on individual outcomes, and the exclusion of unsuccessful participants.
  3. It's frustrating to see that the failure of weight loss interventions has been consistently validated since the 1950s, yet the weight loss industry continues to grow, offering dangerous and expensive solutions.

Bridging the Gap: Navigating the Curse of Knowledge in UX

UX Psychology • 178 implied HN points • 23 Feb 24
🕹 Technology Research
  1. The curse of knowledge is a cognitive bias where experts find it hard to imagine not knowing what they know, this can impact different fields, including UX.
  2. Cognitive biases like the curse of knowledge stem from how our brains process and remember information, leading to errors in thinking and decision-making.
  3. To overcome the curse of knowledge, involve users in the design process, seek feedback from colleagues, and communicate in plain language to ensure better collaboration and user experience.

Interracial Dating Paper FAQ

sebjenseb • 196 implied HN points • 10 Feb 24
🔬 Science Research
  1. Assortative mating occurs between races, with individuals who date outside their race being more similar to each other in terms of intelligence, height, and risk-taking behaviors.
  2. Current literature suggests that interracial relationships may have a higher likelihood of ending or experiencing domestic violence issues, and mixed-race children might be more prone to mental/behavioral problems, possibly due to self-selection rather than social factors.
  3. Attractiveness was a weak predictor of interracial dating across all races, indicating that mate value or race exchanges based on mate value were not significant factors in interracial dating.

Space Ambition: 2023 Year in Review

Space Ambition • 259 implied HN points • 30 Dec 23
🕹 Technology Research
  1. The subscriber count grew seven times this year, increasing from 1,500 to over 9,800. This shows strong interest in spacetech content across many countries.
  2. The blog published 52 posts, with an impressive average of 5,000 to 7,500 views each. Popular sections included interviews with industry leaders and insights into various countries' space tech ecosystems.
  3. The team was active in the spacetech community, hosting events and mentoring startups, while also achieving a successful satellite launch. They valued the contributions of volunteers and the support from recognized blogs.