flâneur

Charles Liu

6 followers · 6 following · 1147 views

on the atlas — 15

highlights — 24

  • The number one thing you can do to motivate cracked people? Find them more cracked people to work with!
    Hiring (and managing) cracked engineers - PostHog
  • From chatbot to agent-coworker What could ambitious unhobbling over the coming years look like? The way I think about it, there are three key ingredients: 1. Solving the “onboarding problem” GPT-4 has the raw smarts to do a decent chunk of many people’s jobs, but it’s sort of like a smart new hire that just showed up 5 minutes ago: it doesn’t have any relevant context, hasn’t read the company docs or Slack history or had conversations with members of the team, or spent any time understanding the company-internal codebase. A smart new hire isn’t that useful 5 minutes after arriving—but they are…
    I. From GPT-4 to AGI: Counting the OOMs - SITUATIONAL AWARENESS
  • Except you probably don’t get to the top 1% of your field by taking a steady, linear path. More often, it’s the result of extreme practice for a short period of time, followed by consistency and creativity.
    ️ Learning in Ways That Dont Scale
  • The hard work you put in during your challenge turns out to be a pretty fantastic investment. You learned the rules of the game. Now you know which ones to break. “In order to scale, you have to first do things that don’t scale at all.” — Reid Hoffman This is where a lot of successful people stand right now. They worked their ass off early on to get to know their field or craft, and are now able to scale back their input while maintaining a certain level of performance or output.
    ️ Learning in Ways That Dont Scale
  • When you don’t want to do something, interrogate the reason.
    27 Principles for 27 Years
  • Both in business and life, double down on what works.
    27 Principles for 27 Years
  • Work backwards from the outcomes you want to see.
    27 Principles for 27 Years
  • Learning new things rarely disappoints as a source of fulfillment. ⭐️
    27 Principles for 27 Years
  • Don't latch your identify onto outcomes. They won't last forever. ⭐️
    27 Principles for 27 Years
  • 1. Actually do the goddamn practice 2. The feedback loop is the primary product.
    Feedbackloop-first Rationality — LessWrong
  • Once you know what problems you should be solving, you need to take action. This could mean taking an “Ask for forgiveness, not permission” approach, depending on how your organization functions. It takes bravery to do this, but like anything else, it's a muscle that you can build up and improve over time. Don't be afraid to make the first move, the repercussions aren't as serious as you think they are.
    Barrels and Ammunition
  • Let's quickly recap. In order to become a barrel in your organization, you should work on mastering each of the following steps: Understand: Develop a mental model of the problem you're solving Ideate: Think deeply about the problem and how to solve it Take initiative: Create convincing proof of concepts for your ideas Recruit others: Build relationships to bring in teammates to help Deliver results: Level up your skills. Ship work that turns heads.
    Barrels and Ammunition
  • Keeping things simple can be very helpful, especially from a model perspective. It can be attractive to use multiple model architectures and complex prompt optimization pipelines - and often they can net you an additional percent or two in accuracy. However, keep an eye on the tradeoff you might be making in terms of complexity and maintainability. The more black-box ML that you add to a pipeline, the harder it becomes to upgrade, maintain and control. When building a new system, start from the query in. Understand the class of questions your system will be expected to answer, and line up expe…
    Better RAG 3: The text is your friend
  • This isn't a new concept - one of the most successful RAG techniques has been HyDE - Hypothetical Document Embeddings, which tries to generate potential (fake) answers to the question, in the hopes that the embeddings of these answers match real answers more closely. You can also do the inverse. In WalkingRAG, one additional step we perform at ingest is to generate hypothetical questions that are answered by each section of a document. The embeddings of these questions are far closer to the actual user question than the source data - it's also a helpful multi-modal step, to generate text quest…
    Better RAG 3: The text is your friend
  • unless the tide turns soon, the Internet I fell in love with may cease to exist, and in its place, we will have something closer to a souped-up version of TV – focused largely on passive consumption, with much less opportunity for active participation and genuine human connection
    Omegle
  • I’ve done my best to weather the attacks, with the interests of Omegle’s users – and the broader principle – in mind. If something as simple as meeting random new people is forbidden, what’s next? That is far and away removed from anything that could be considered a reasonable compromise of the principle I outlined. Analogies are a limited tool, but a physical-world analogy might be shutting down Central Park because crime occurs there – or perhaps more provocatively, destroying the universe because it contains evil. A healthy, free society cannot endure when we are collectively afraid of each…
    Omegle
  • Fear can be a valuable tool, guiding us away from danger. However, fear can also be a mental cage that keeps us from all of the things that make life worth living. Individuals and families must be allowed to strike the right balance for themselves, based on their own unique circumstances and needs. A world of mandatory fear is a world ruled by fear – a dark place indeed.
    Omegle
  • In recent years, it seems like the whole world has become more ornery. Maybe that has something to do with the pandemic, or with political disagreements. Whatever the reason, people have become faster to attack, and slower to recognize each other’s shared humanity.
    Omegle
  • PCA rewrites the music so that fewer performers can play the same song.
    Principal Component Analysis (PCA) | by Shaw Talebi | Towards Data Science
  • The biggest benefit, however, comes from how The Transformer lends itself to parallelization.
    The Illustrated Transformer – Jay Alammar – Visualizing machine learning one concept at a time.
  • The context vector turned out to be a bottleneck for these types of models. It made it challenging for the models to deal with long sentences. A solution was proposed in Bahdanau et al., 2014 and Luong et al., 2015. These papers introduced and refined a technique called “Attention”, which highly improved the quality of machine translation systems.
    Visualizing A Neural Machine Translation Model (Mechanics of Seq2seq Models With Attention) – Jay Alammar – Visualizing machine learning one concept at a time.
  • adopting masked signal modeling techniques, which have been proven effective in capturing contextual information from noisy and variable data [18, 7], represents a promis- ing avenue for deriving meaningful contextual knowledge from large-scale noisy EEG dat
    2306.16934.pdf
  • The temporal resolution of EEG is high, meaning that it can capture rapid changes in brain activity that occur on the order of milliseconds. However, the spatial resolution of EEG is low, meaning that it is difficult to precisely lo- calize the source of the activity within the brain. Secondly, EEG signals are highly variable, influenced by factors such as age, sleep, and cognitive state. Finally, EEG data is of- ten noisy, and requires careful processing and analysis to extract meaningful information
    2306.16934.pdf
  • Not by way of ideological stubbornness, but by being engrossed enough with your own thing to develop mental models about how things should work
    how to coparent baby ideas - by Anson Yu