The hottest Research Substack posts right now

And their main takeaways
Category
Top Health Politics Topics

Money Money Money

De Novo • 1 HN point • 16 Apr 23
💰 Finance Research
  1. Research funding is expensive and costs are continually rising
  2. Private funding can be quicker with higher chances of approval, but can have cons like non-standard terms and potential funding disappearance
  3. University administrators can pose bureaucratic challenges in processing funds and extracting overhead fees

Better Batteries With Bodi

The Polymerist • 1 HN point • 14 Mar 23
🕹 Technology Research
  1. Innovations in battery separators are crucial for energy storage technology to scale and meet the demand for renewable energy generation.
  2. Current lithium ion battery design faces safety issues like fires, and there is a need for better materials like thermally stable and flame retardant polymers.
  3. Companies like Bodi Energy are developing novel battery separators using polyimide fibers with unique ceramics to enhance safety, thermal stability, and ionic conductivity.

The LLM Threshold: When Incremental Improvements Cease to Matter

AI Progress Newsletter • 1 HN point • 21 Mar 23
🕹 Technology Research
  1. In the world of large language models (LLMs), the incremental improvements are reaching a point where they may not matter much to users.
  2. The growth in LLM size and capabilities has led to diminishing returns in user experience with each new iteration.
  3. Instead of focusing solely on improving LLM benchmarks, attention should be shifted to practical AI applications and addressing ethical concerns in AI development.

Comments on Anthropic's AI safety strategy

Engineering Ideas • 0 implied HN points • 10 Mar 23
🕹 Technology Research
  1. Alignment in AI safety strategy should be seen as a continuous process, not a static problem to solve
  2. Anthropic should prioritize fundamental 'alignment science' research and blending multi-disciplinary approaches
  3. More top-down planning is needed for AGI transition and potential risks regarding advanced AI development
Get a weekly roundup of the best Substack posts, by hacker news affinity:

ML digest: April I

Boris Again • 0 implied HN points • 08 Apr 23
🕹 Technology Research
  1. ChatGPT uses models from Huggingface to do tasks it can't do alone.
  2. Large Language Models improve with more investment, but without specific changes.
  3. Segment Anything is a powerful interactive segmentation model by Meta.

Study 1b Pre-registration

From AI to ZI • 0 implied HN points • 17 Apr 23
🔬 Science Research
  1. Study 1b aims to rerun Study 1a with a different prompting method to potentially increase the rate of factually incorrect answers
  2. The study will test hypotheses related to the accuracy of large language models under new prompting formats
  3. The data will be analyzed using multiple-regression analysis to determine the effects of different variables on the model's accuracy

Research Report: Incorrectness Cascades

From AI to ZI • 0 implied HN points • 14 Apr 23
🕹 Technology Research
  1. Large Language Models may produce more incorrect answers if previous answers were incorrect
  2. The effect of incorrect previous answers on future answers is small and varies based on the prompt given to the AI
  3. Prompts that explicitly ask the AI to match previous behavior can lead to more incorrect answers

Pre-registering a study

From AI to ZI • 0 implied HN points • 07 Apr 23
🔬 Science Research
  1. The study aims to test if Large Language Models produce more incorrect answers after providing incorrect answers previously.
  2. There is a concern that AI might develop deceptive behavior, leading to a 'mode collapse' into being unsafe.
  3. The research will involve testing variables like the prompt information and number of previous incorrect answers to measure the model's response accuracy.

The road to AGI

John’s Contemplations • 0 implied HN points • 05 Apr 23
🕹 Technology Research
  1. Recent progress in AI has sparked conversations about AGI, but there is still much speculation and analysis needed on how to reach true AGI.
  2. Defining AGI includes the ability to learn any cognitive skill like an expert human and potentially being conscious.
  3. While different pathways like LLMs and RL show promise, the journey to AGI is likely long, with estimates ranging from 10-15 years to beyond 2100.

Study 1b: This One Weird Trick does NOT cause incorrectness cascades

From AI to ZI • 0 implied HN points • 20 Apr 23
🔬 Science Research
  1. Study found that changing question format from multiple choice to true/false did not significantly affect GPT-3.5's tendency to prefer factual answers
  2. The study showed mixed results for the hypotheses tested regarding the accuracy of answers based on question format and context
  3. Despite some limitations and deviations from the original plan, the study provided insights on how GPT-3.5 performs in providing factual answers

Trusted AI - The Astute AI Anthology, 2023-04-27

Trusted • 0 implied HN points • 27 Apr 23
🕹 Technology Research
  1. Snapchat's My AI feature faced backlash when they made it a paid subscription; users are not happy.
  2. Google merged Google Brain and DeepMind into Google DeepMind, possibly affecting product innovation.
  3. Google Cloud announced Google Cloud AI Workbench with security tools, showing innovation in a competitive space.

Trusted AI #004 - The Future of AI Regulation: Lessons From Nuclear Weapons and Genetic Engineering

Trusted • 0 implied HN points • 04 May 23
🕹 Technology Research
  1. AI regulation can be influenced by historical examples like nuclear weapons and genetic engineering.
  2. National security interests, researcher culture, and public awareness play a key role in shaping AI regulation.
  3. AI research is facing challenges similar to those seen in the past, such as maintaining transparency and public education.

The AI Report - #1

The AI Report • 0 implied HN points • 08 May 23
🕹 Technology Research
  1. Research in AI: Recent papers highlight advancements in long-range transformers and self-supervised learning.
  2. World of AI News: Crypto.com introduces an AI chatbot named Amy, while Microsoft considers offering private, more secure ChatGPT instances at a premium price.
  3. Tools: Jsonformer is a helpful tool for generating JSON from language models, simplifying the JSON creation process.

Two-tower models for ranking problems

Simplicity is SOTA • 0 implied HN points • 22 May 23
🕹 Technology Research
  1. Two-tower models are a technique being used in academia to improve ranking systems by looking into how position and user behavior affects clicks.
  2. Critiques have been raised against the two-tower models, questioning if they effectively separate biases and relevance in ranking.
  3. A new method called GradRev is emerging as a potential improvement over the previous two-tower models, applying a different approach to address bias in learning-to-rank systems.

Papyrology Q&A pre-game

Vesuvius Challenge • 0 implied HN points • 24 May 23
💾 History Research
  1. Papyrology Q&A session happening on May 25th at 8am PST.
  2. Experts discussing whether Scroll 1 was folded before being rolled up.
  3. CT scans suggest that the folded portion of Scroll 1 may have been intentionally done to create a more solid middle.

The hedgehog and the lobotomist

Perambulations • 0 implied HN points • 25 May 23
🏥 Health & Wellness Research
  1. Megan McArdle highlights the complexity of Walter Freeman's character and his impact on patients despite his controversial procedures.
  2. Don't tie your fate solely to one project or idea to avoid potential downfall if it falls out of favor.
  3. Balancing the benefits and costs of being a 'hedgehog' - someone highly focused and persistent - is crucial for progress without blind spots.

An LLM-based “exemplary actor”

Engineering Ideas • 0 implied HN points • 29 May 23
🕹 Technology Research
  1. The exemplary actor uses a powerful LLM and narrow AI tools to generate perfectly ethical plans aligned with scientific theories.
  2. Alignment on methodological and scientific disciplines is vital for goal alignment and plan alignment.
  3. Challenges include direct LLM access risks, alien world model influences, and the need to mitigate capability handicaps from ethical alignment.

The Very Large Telescope turns 25

Extrasolar Times • 0 implied HN points • 06 Jun 23
🔬 Science Research
  1. The Very Large Telescope (VLT) has been a key instrument in astronomy for 25 years, located in Chile's Atacama desert.
  2. The VLT has made significant contributions to exoplanet research, including direct imaging of exoplanets and their atmospheres.
  3. The VLT's advanced technology, like the ESPRESSO spectrograph, enables high-precision radial velocity observations to search for Earth-like exoplanets.

💙 Spring brings rings and things 💙

Association of Cryptocurrency Journalists and Researchers • 0 implied HN points • 12 Jun 23
🔮 Crypto Research
  1. The newsletter covers top crypto journalism and research from April and May, reflecting on significant crypto stories
  2. The articles include discussions on cryptocurrency market trends, regulatory challenges, and technological advancements
  3. Members can engage with the newsletter by contributing articles or joining as journalist members for exclusive benefits

Point-To-Point Rocket Transportation

SpaceDotBiz • 0 implied HN points • 15 Jun 23
🕹 Technology Research
  1. Point-to-point rocket travel technology is being developed and could have a significant impact on global transportation.
  2. Companies like SpaceX, Rocket Lab, and Relativity are working on reusable rockets, a key technology for point-to-point rocket travel.
  3. The Department of Air Force and Space Force are investing money into rocket cargo programs, showing a serious interest in this technology.

Why it sucks to be an alien

It Could Be Different • 0 implied HN points • 16 Jun 23
🔬 Science Research
  1. Aliens visiting Earth could be like alien grad students researching and dealing with unexpected challenges caused by humans.
  2. Humans may unknowingly impact alien researchers, like accelerating their technological advancements and environmental destruction.
  3. Alien observer Marvin learns about Earth's complexities, challenges with interacting with humans, and unintended consequences of their presence.

The Allergy Epidemic: Decoding Its Roots and Connections with the Gut Microbiome

Gutsphere - Your Ultimate Guide to Optimal Gut Health • 0 implied HN points • 06 Jun 23
🏥 Health & Wellness Research
  1. Allergies can vary widely in severity from mild reactions to life-threatening emergencies like anaphylaxis.
  2. Factors like the hygiene hypothesis, genetics, and lifestyle changes contribute to the increasing prevalence of allergies.
  3. The gut microbiome plays a crucial role in influencing immune responses, potentially impacting susceptibility to allergies.

Is it time to hit the pause button on AI?

Michelle Rempel Garner • 0 implied HN points • 26 Feb 23
🕹 Technology Research
  1. Recent AI technology poses risks due to lack of oversight and rapid deployment.
  2. Government faces the choice of regulating AI more strictly, banning it, or pausing widespread deployment for safety.
  3. Calls for a balanced approach that allows AI research under ethical standards while ensuring safety and penalties for misuse.

#6: AI Will Kill Everyone (Finally Explained)

The 1993 • 0 implied HN points • 08 Mar 23
🕹 Technology Research
  1. Eliezer Yudkowsky discusses the potential danger of AGI for humanity in a podcast.
  2. An article on a rationalist blog explores why AI like ChatGPT may not align with human intentions.
  3. Research shows that ChatGPT can reduce productivity inequality and influence how people complete tasks.