bartwr · HackerLangs

2 か月前議論

I have most likely outdated info, I left Google Research 4y ago. Back then, available TPU instances were plenty and GPU scarce. Nobody wanted to mess with an immature crashing compiler and very steep performance cliffs (performance was excellent only if you stayed within the guardrails, and being outside was supported and not even resulting in a warning - as it was so common in code). But I believe most of it has changed for the better for TPUs.

bartwr

3 か月前議論

My experience cycling regularly in NYC: bike lanes separated by curb, stoppers, or poles are more dangerous as cars stop at their entrances/exits and I am literally trapped or cannot enter them before/after an intersection. I'm not against them in principle, but without extremely strict enforcent of laws (let's say a ticket 5% of someone's annual income and a loss of DL on a repeated offense - this stuff endangers people's lives), they are sadly counterproductive. :(

bartwr

3 か月前議論

I love using AI tools and they are changing my work and life in amazing ways. I cannot imagine going back. And yet, I am more concerned about the social damage due to their widespread use and the amounts of slop they generate. Just this week: - There was an article about a news company faking polls by asking LLMs for answers. - My wife told me that she stopped watching any funny pet videos because 99% now is AI slop - start normal, but then turn into someone's slop idea. - A friend told me their big tech company uses AI-generated metrics as part of performance evaluation. Nobody checks them. - Another friend told me their big tech company requires engineers to use AI-generated commit messages with terrible signal-to-noise ratio and making version control and history useless for engineers. But directors and PMs love them, they are so descriptive! - My neighbor uses LLMs to create some neighbor meeting plans/agendas, plausibly looking PDFs citing contractors etc. It's impossible to read through it, mixed hallucinations and real information, all wrapped in thousands of slop words. What is real and what made up? I'll spend 10x more time double guessing. - Encountering more and more articles and general "content" that is AI generated and looks ok at the first glance, but slop upon inspections. Why would I read LLMs output on a webpage with ads, if I can ask it myself and get better, personal answers and style?

And I am not even talking here about other ethical issues, training data, less junior job positions, job replacement of journalists with LLM-equipeed contractors, etc.

LLMs make my personal and work life so much better, but social life unbearable. Is it worth the trade-off? I guess it doesn't matter at this point.

bartwr

5 か月前議論

It's hilarious that people describe their anecdotal experience of being in a calorie deficit as a proof that "intermittent fasting works". This was never the question, but whether intermittent fasting brings additional weight loss benefits as compared to calorie deficit with frequent meals.

My anecdotal experience from 20y of bodybuilding and doing ~3 cuts a year: for cutting, I tried IF, 6 meals a day, low fat, low carb, high fat true keto, balanced... everything works. And works equally well - this is backed by numerous studies. The only difference is the impact on health parameters (different will get worse on low fat vs high fat), satiety, and how easy it is for someone to sustain the diet and stay in a deficit. This will depend on the lifestyle and personal preferences. So my preferred way to cut is high protein, low carb, essential fats, a ton of fiber. When building muscle I go high everything but balanced.

Anything else and more is sectarianism and people bragging about their choices not having verified their true claimed efficacy or benefits.

bartwr

3 年前議論

Sadly things take so long because democratic rule of law law "has to" work slowly (you would not want "innocent" businesses closed overnight or citizens arrested because of allegations). It's not great, obviously, and both false positives and negatives happen.

The issue is that people and companies exploit it, "better ask for forgiveness than permission", and after years they clean up evidence, their CEOs say they "don't remember", and even if they get fined - it's too late and will not save destroyed, monopolized market. All those gig economy "disruptors" are the worst, but probably all the biggest companies do this.

I wish that there could be a possibility of a retribution payment - a monopolistic payment destroyed an $xyz bln competitor? They need to pay this amount of money, maybe to some publicly owned alternative/open source competition if it's too late.

bartwr

3 年前議論

One aspect of SBF's downfall that is not discussed enough, IMO: amphetamines (even taken for medical reasons/ADHD) release a ton of dopamine, seek more release, and make you do some weird stuff and seek risky behaviors (and reinforce them through reward system).

I recall his interviews and annoyed tone at uncomfortable questions - made me laugh, just like people on speed.

And absolutely avoid selegiline for "productivity", especially with amphetamines, this will turn anyone into a pathological compulsive person (e.g. a gambler).

コメント