Varun Shenoy
102 followers · 36 following · 15419 views
on the atlas · 308
- Fontshare: Quality Fonts. Free.9 savers
- School Is Not Enough25 savers
- Junto (club)5 savers
- The best general advice on earth « the jsomers.net blog21 savers
- Barnum effect7 savers
- 406-706-Looking-Back-Moving-Forward-Jessica-Handler.pdf5 savers
- Maker's Schedule, Manager's Schedule10 savers
- How I ship projects at big tech companies | sean goedecke8 savers
- How to Twitter Successfully | near.blog11 savers
- Do Ten Times as Much - by Bryan Caplan - Bet On It11 savers
- themundanityofexcellence.pdf6 savers
- Market Ending Moves - by Elad Gil - Elad Blog4 savers
- How to write complex software3 savers
- DOGE Days8 savers
- How I Build — Adam Montgomery2 savers
- Beware the Casual Polymath | Applied Divinity Studies9 savers
- Spaced Repetition Systems Have Gotten Way Better | Domenic Denicola8 savers
- Find Your People - Jessica Livingston12 savers
- Notes on Superhuman’s Acquisition Loops - kwokchain2 savers
- Letter #159: Richard Rainwater (1997) - by Kevin Gee1 savers
- Don't Call Yourself A Programmer, And Other Career Advice | Kalzumeus Software1 savers
- Reading as the root of career planning - Commoncog1 savers
- Everyone is Talking Their Book3 savers
- 28 slightly rude notes on writing - by Adam Mastroianni6 savers
- M&A Canon | Letters of Intent6 savers
- Career Path Liquidity Traditional Linear Paths vs Non-Linear Paths5 savers
- Robot Dexterity Still Seems Hard - by Brian Potter2 savers
- The Road to Self-Renewal. This is a speech by John Gardner… | by Junto | Junto | Medium2 savers
- Most AI value will come from broad automation, not from R&D | Epoch AI11 savers
- You And Your Research · Gwern.net13 savers
- The Architecture of Open Source Applications (Volume 1)Python Packaging1 savers
- The C10K problem1 savers
- The Architecture of Open Source Applications (Volume 2)nginx2 savers
- Sashimi Quality – Truefish1 savers
- The Deployment Age | Reaction Wheel2 savers
- lehman.pdf1 savers
- Out of the Tar Pit1 savers
- Management Learnings with Jamie Dimon I JPMorganChase - YouTube1 savers
- S**t Umbrella – Zach Perret2 savers
- How I've run major projects | benkuhn.net53 savers
- You might not need Redis - viblo.se1 savers
- The Economy Is Stupid - by J. Daniel Sawyer2 savers
- The momentum of mission. - by Frederik Gieschen2 savers
- Ask for no, don’t ask for yes – Dan Moore!4 savers
- ALL React Hooks Explained in 12 Minutes - YouTube2 savers
- Lessons from Peter Thiel | Posts | 8VC15 savers
- Palantir: The Founder Foundry - by Apoorv Agrawal4 savers
- How To Ask Questions The Smart Way12 savers
- Databases in 2024: A Year in Review // Blog // Andy Pavlo - Carnegie Mellon University3 savers
- Fantastic Learning Resources2 savers
- minjunes.ai/posts/sleep/index.html4 savers
- Spaced Repetition for Efficient Learning · Gwern.net4 savers
- ProofsArgsAndZK.pdf3 savers
- Sorry, that isn't an FDE - by Ted Mabrey - Ted’s Substack2 savers
- 27 Pieces of Advice on my 27th Birthday - Jackson Dahl13 savers
- Rostra5 savers
- How to De-Risk a Startup - by Leo Polovets - Coding VC2 savers
- How and why we built our startup around small teams2 savers
- I Don’t Want to Be an Internet Person9 savers
- The Secret History And Strange Future Of Charisma3 savers
- The Waluigi Effect (mega-post) - LessWrong16 savers
- Codus Operandi11 savers
- True Product Market Fit is a Minimum Viable Company | by Ann Miura-Ko | The Startup | Medium5 savers
- Mamba Explained2 savers
- You're probably wrong about how things have changed3 savers
- How to keep cakes moist and cause the greatest tragedies of the 20th century3 savers
- Aim, fire, scan: the 80/20 of executing on big projects12 savers
- LLMs shouldn’t write SQL - by Benn Stancil - benn.substack2 savers
- Linear/Fully-Connected Layers User's Guide :: NVIDIA Deep Learning Performance Documentation1 savers
- E.W.Dijkstra Archive: On the foolishness of "natural language programming". (EWD 667)2 savers
- The Top Idea in Your Mind6 savers
- Some Intuition on Attention and the Transformer3 savers
- LLM Inference Performance Engineering: Best Practices | Databricks Blog6 savers
- The Load-Bearing Relationship17 savers
- Nvidia Envy: understanding the GPU gold rush2 savers
- The Four Burners Theory: The Downside of Work-Life Balance2 savers
- Exceptional5 savers
- You Can’t Trust the AI Hype4 savers
- Uncensored Models6 savers
- Equity Guide for Employees at Fast-growing Companies2 savers
- Timing Technology: Lessons From The Media Lab · Gwern.net3 savers
- Examples of barbell strategies - by Dwarkesh Patel4 savers
- If you’ve raised venture capital, you have to pay yourself | TechCrunch2 savers
- The Cathedral and the Bazaar2 savers
- Existential risk, AI, and the inevitable turn in human history - Marginal REVOLUTION11 savers
- Tailwind CSS Color Generator - UI Examples & WCAG Contrast Ratios7 savers
- Startups are an act of desperation - by Elad Gil5 savers
- LLM Powered Assistants for Complex Interfaces - Nick Arner2 savers
- I felt obliged to comment because I feel I know what you are talking about and I... | Hacker News4 savers
- The magic of traveling alone - Yuval Aizenman2 savers
- How to Get Startup Ideas4 savers
- Steve Job's Life Philosophy - by David Senra5 savers
- Why didn't DeepMind build GPT3? - by Jonathan Godwin4 savers
- leaving | MIT Admissions2 savers
- H3: Language Modeling with State Space Models and (Almost) No Attention · Hazy Research5 savers
- My 8 Best Techniques for Evaluating Character5 savers
- Using ZK Proofs to Fight Disinformation | by Dan Boneh | Sep, 2022 | Medium4 savers
- You'll forget most of what you learn. What should you do about that?22 savers
- Why the Age of American Progress Ended - The Atlantic2 savers
- AI: Markets for Lemons, and the Great Logging Off3 savers
highlights · 2188
Current generation LLMs are way too comment happy. Comments should be reserved for decisions that are not intuitively obvious from the code. These include decisions to omit more robust handling (i.e. TODOs), and any algorithmic choices that would make a good leetcode question.
How I Build — Adam MontgomeryCurrent generation LLMs have a bad habit of catching exceptions, logging them, and returning a default value or object. That usually compounds the issue. Good software handles the exceptions it anticipates, and loudly reports the ones it doesn't. A 500 Internal Server Error response is loud and clear. A 200 OK with an empty list is just confusing.
How I Build — Adam MontgomeryRichard Rainwater: Well, probably the most important thing is that capitalism is the only game. It's the best game. We pioneered it in this country. There's more information about capitalism in this country than the rest of the world put together. The great asset allocators that run mutual funds are available to you. It's a very easy and simple thing to basically play that game, and you ought to be doing it.
Letter #159: Richard Rainwater (1997) - by Kevin GeeWell, I hope it goes to 14 or even lower, because that'll give us more opportunities, hopefully, to continue to expand our businesses at low prices.
Letter #159: Richard Rainwater (1997) - by Kevin GeeAdd revenue. Reduce costs. Those are your only goals.
Don't Call Yourself A Programmer, And Other Career Advice | Kalzumeus SoftwarePersonally, I think the reason is far more sinister: making art is painful because it forces the mind to do something it’s not meant to do. If you really want to get that sentence right, if you want that perfect brush stroke or that exquisite shot, then you have to squeeze your neurons until they scream. That level of precision is simply unnatural.
28 slightly rude notes on writing - by Adam MastroianniRoboticist Rodney Brooks recently predicted that deployable humanoid robot dexterity will be “pathetic” compared to humans beyond 2036.
Robot Dexterity Still Seems Hard - by Brian PotterFirst, from a business perspective, we suspect that for the foreseeable future, it will be more profitable for AI labs to focus on trying to automate ordinary work tasks—such as creating computer-use agents that can competently browse the internet, operate commercial software, and perform standard white collar job tasks—rather than focusing on developing “Nobel Laureate”-level reasoning models that are capable of narrowly assisting researchers in fields like biology and medicine.
Most AI value will come from broad automation, not from R&D | Epoch AIKey distribution deal. IBM distributing Microsoft’s early O/S, or Yahoo! distributing Google were two “king making” moves in an industry.
Market Ending Moves - by Elad Gil - Elad BlogAnd the job of an employee is to don a shit-proof jacket, and find a way to get your work done.
S**t Umbrella – Zach PerretIn all likelihood, they won’t take the same path, and the presence of others in your maze means you might be onto something. Your real competition – and what you should worry about – is the years you could waste going down the wrong path.
cdixon | The idea mazeOne habit I like is to ask for no, not yes.
Ask for no, don’t ask for yes – Dan Moore!never, never let the universe smooth you into your surroundings. It remains Day 1.
2020 Letter to ShareholdersIn what ways does the world pull at you in an attempt to make you normal? How much work does it take to maintain your distinctiveness?
2020 Letter to ShareholdersWhen we die the work stops, the temperature differential starts to disappear, and we end up the same temperature as our surroundings.
2020 Letter to ShareholdersThe reality is that these relationships are enforced, by profit-seeking intermediaries.
Making Markets in Time - by Abraham Thomas - PivotalIn someone else’s hands, the work could have turned out average. It’s living the answer to the question, “If you weren’t there, how would this have turned out differently?”
Turning a yellow spot into the sun - Wes Kao's Newsletter7. For written communication, use “main point above, context below.”
How to be concise - Wes Kao's Newsletter3. Avoid explaining events chronologically.
How to be concise - Wes Kao's NewsletterYou: “We may need to give even more guidance to new hires.”
How to give a senior leader feedback (without getting fired)When you haven’t yet traded who you could be in for who you are, for who you have decided to become.
on trading potential for something actual - by Isabelthe very compelling sense of potential that everything has when you are still fairly ‘undifferentiated’ in your life.
on trading potential for something actual - by IsabelDo not mock things that don't do IO. Why would you? You have the real implementation!
How to write complex softwareThe only thing you want to consider producing separate "real" mocks for is the IO layer. You want to do this for tests, because you probably want your tests to be fast and deterministic.
How to write complex softwareThe right way to do it is to code the simple version first. Code it that way before you have implemented the rate_limiter or the article_store or whatever. Just imagine they exist. Code as if the perfect API to implement this layer already existed.
How to write complex softwareThe method I propose is the exact opposite. You start coding at the top of the stack, letting each layer define the API of the layer below it. Only once that layer's implementation is beautiful — because the stubbed out API of the layer down is just what it needs — do you implement the layer down.
How to write complex softwareWhen coding a layer, define the API of the next layer down to be the perfect domain-specific language to implement the current layer in
How to write complex softwareWorkflows are systems where LLMs and tools are orchestrated through predefined code paths. Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.
Building effective agents \ AnthropicYou can tell that Snowflake cares most about taking a shot at Databricks because their announcement shows other LLMs doing better than them (e.g., Llama3), but they highlight how they are better than DBRX. One AI researcher was confused about why Snowflake focused so much on DBRX in their analysis and not the other models; this person does not know how much blood these two database rivals have spilled.
Databases in 2024: A Year in Review // Blog // Andy Pavlo - Carnegie Mellon UniversityMany people believe there can be equal co-founders who got started ‘together’, but Koenig has found that in every case there is one single source.
How initiatives start | work with sourceExceptions exist for exceptional situations: events that are not a part of normal execution.
Robust exception handling - Eli Bendersky's websiteIt prevents ergonomic function composition
Rust Solves The Issues With Exceptions · Dmitrii Aleksandrovavoid affectation in every way possible as though it were some rough and dangerous reef;
SprezzaturaAnd that seemed like a great opportunity, even though I didn’t really want to do investing anymore. I thought I could learn a ton from a firm like Sequoia especially doing early-stage investing.
[Transcript] Building a Customer-Centric Culture with David Vélez, Founder and CEO of Nubank | by Miguel Armaza | MediumEither that, or admit that you’ve got higher priorities.
Do Ten Times as Much - by Bryan Caplan - Bet On ItContractors spend their days interfacing with the ServiceTitan platform across what we believe to be the five most business-critical functions, or the “core centers of gravity,” inside a trades business: CRM (customer relationship management, including sales enablement, marketing automation and customer service), FSM (field service management, including scheduling and dispatching), ERP (enterprise resource planning, including inventory), HCM (human capital management, including compensation and payroll integration) and FinTech (including payments and third-party consumer financing).
ServiceTitan: Benchmarking the S1 Data - by Jamin BallWhen you begin a new skill, cognitive load will be at its highest.
The Biohacker's Guide to Rapid Skill Acquisition: Using Pharmacology and Neuropsychology to Build SkillsThe Cerebras guy also mentioned that a lot of their recent customer interest has been from companies upset about the Nvidia change in terms of service (the one that forces companies away from consumer grade GPUs to enterprise Pascals / Voltas)
OpenAI Email Archives (from Musk v. Altman) — LessWrongIf we do this really well, the transportation industry is large enough that we could increase Tesla's market cap to high O(~100K), and use that revenue to fund the AI work at the appropriate scale.
OpenAI Email Archives (from Musk v. Altman) — LessWrongHe did mention that Holden was irked by the move to for-profit and potentially offered more substantial amount of money if OpenAI stayed a non-profit, but hasn't firmly committed.
OpenAI Email Archives (from Musk v. Altman) — LessWrongEither go do something on your own or continue with OpenAI as a nonprofit.
OpenAI Email Archives (from Musk v. Altman) — LessWrongIs AGI truly your primary motivation? How does it connect to your political goals? How has your thought process changed over time?
OpenAI Email Archives (from Musk v. Altman) — LessWrongWe don't understand why the CEO title is so important to you. Your stated reasons have changed, and it's hard to really understand what's driving it.
OpenAI Email Archives (from Musk v. Altman) — LessWrongIn the event we decide to buy Cerebras, my strong sense is that it'll be done through Tesla.
OpenAI Email Archives (from Musk v. Altman) — LessWrongThe current structure provides you with a path where you end up with unilateral absolute control over the AGI.
OpenAI Email Archives (from Musk v. Altman) — LessWrong-$200M-$1B initial raise.
OpenAI Email Archives (from Musk v. Altman) — LessWrong-One concern they had was the proposed employee pool was too small.
OpenAI Email Archives (from Musk v. Altman) — LessWrong-Greg noted that Ilya in some ways has contributed millions by leaving his earning potential on the table at Google.
OpenAI Email Archives (from Musk v. Altman) — LessWrongFine by me if they don't use this in active messaging. Would be worth way more than $50M not to seem like Microsoft's marketing bitch.
OpenAI Email Archives (from Musk v. Altman) — LessWrongThis actually made me feel nauseous. It sucks and is exactly what I would expect from them. Evaluation, Evangelization, and Usage of CNTK v2, Azure Batch and HD-Insight: OpenAI will evaluate CNTK v2, Azure Batch, and HD-Insight for their research, provide feedback on how Microsoft can improve these products. OpenAI will work with Microsoft to evangelize these products to their research and developer ecosystems, and evangelize Microsoft Azure as their preferred public cloud provider. At their sole discretion, and as it makes sense for their research, OpenAI will adopt these products Let’s just …
OpenAI Email Archives (from Musk v. Altman) — LessWrong