Hacker News Reader

I'll add

- Flip the back and recent apps button in three button navigation. It's such a weird decision - like only left handed people should be able to comfortably use three button nav. Literally hurt my thumbs to use the phone.

- The quick settings panel is just abysmal compared to Samsung with massive pill buttons that have virtually nothing in them.

odneson Ask HN: Working in tech for climate?

Surprised no one has said the following; study AI and try make a meaningful contribution to the field. It may sound trite, but the most likely way we solve the climate crisis is by solving AGI first, hoping it doesn't cause a doomsday scenario itself, and subsequently hoping 'it' will coordinate an effort to reverse the damage we have done.

odneson First images of cerebrospinal fluid in the brain during sleep (2019)

The pre-print say the pulses occur with a frequency of ~0.05 Hz (once every 20s). So at a guess the video is sped up 20x.

odneson Now that machines can learn, can they unlearn?

Both MAS and (the earlier) EWC facilitate continual learning through passing a bunch of samples through the network and collecting gradients to determine which weights are 'important'. Future weight changes are then regularised by these importance values so that the network retains its ability on past tasks. EWC uses square gradients as importance values, whereas MAS uses absolute gradients... Other than that they're the same lol (I think), how the MAS paper got so many citations I have no idea.

odneson Amazon refused to refund $7k after shipping an empty box instead of a Sony A1

Hmm, maybe I'm wrong then. Though I could imagine that if this happened to you legitimately that you would have some legal recourse to get your money back; you fulfilled your end of the return by posting it, its not your fault the courier screwed up.

odneson Amazon refused to refund $7k after shipping an empty box instead of a Sony A1

Well there is nothing for them to check. From what I've read, they wait two weeks then mark the package as 'lost in transit'.

I think for the last-mile of delivery, couriers rely on the actual address written on the package, not the address that the barcode scans to. I might be wrong though.

odneson Amazon refused to refund $7k after shipping an empty box instead of a Sony A1

This this a clear cut case and amazon should process the refund, however, in general refunding is a hard problem for Amazon to solve due to the sheer volume fraud that occurs.

For anyone unaware, go to telegram and do a public search for 'refunding'. There are hundreds of channels where you can pay someone to refund >£10,000 of stuff for you for a 10% fee. Afaik, the main method that 'refunders' use to defraud amazon is to (1) initiate a return (2) modify the return label so UPS accepts it into their system but (3) deliver it to the wrong address. So it looks to amazon as being successfully delivered to the return warehouse, but what actually got delivered was a brick, to some random house.

I wouldn't be surprised if >1% of macbook and iPhone refunds are fraudulent in this way. Someone in a cybercrime lab they should write a paper about this whole ecosystem because it is a huge black market.

odneson World Now Likely to Hit Watershed 1.5 °C Rise in Next 5 Years

Do you have and references on your last point about algae? I have this (totally unsubstantiated) idea that plundering the oceans of fish along with rising temperatures will result in the oceans sequestering incredible amounts of CO2 via algae blooms. Never found much to read on the topic though - in what ways could this be negative feedback loop?

odneson Tether reserves backed by 2.9% cash

What's weird is that a couple of weeks ago Coinbase launched trading in Tether on Coinbase Pro. You would think this exposes them to much risk. I've been scratching my head about why they did it.

odneson UK businesses caught buying five-star Google reviews

I'm convinced that something similar is happening to Boris Johnson's tweets at the moment. Every time he posts, many of the top responses are exact copies of each other saying something like "the country is behind you, my prime minister <3". It's clear to anyone that these are 'fake', and I'd imagine it has the effect of making people thing that the conservatives are trying to artificially bolster his reputation.

odneson The Pile: An 800GB Dataset of Diverse Text for Language Modeling

How are they sourcing/funding the compute to train these massive models?

odneson Oxford-AstraZeneca coronavirus vaccine approved for use in UK

> An issue that I basically never see come up is - why not administer the vaccine earlier on (pre-approval) to the most vulnerable? if P(death|vaccine) < P(death|covid) then it seems really barbaric not to give it.

Not sure, but I would think it comes down to hospitals not being overrun in the case of a widescale negative response to the vaccine. But since the whole thing is a game of probabilities, I do share your view.

odneson Why traditional reinforcement learning will probably not yield AGI [pdf]

> The idea of using "rewards" as a primary learning mechanism and a path to actual cognition is just wrong, full stop. It's a wrong level of abstraction and is too wasteful in energy spent.

A lot of research in RL is focused on intrinsic motivation and the question of whether we can bootstrap our own 'rewards' from our ability to predict and control the future according to some self-defined goals/hypotheses.

odneson The Nikon Small World Competition

I wonder how plausible it is to genetically engineer algae to sequester more carbon e.g. make them bigger and heavier so they sink to the bottom of the ocean, and also make them reproduce faster.

Does anyone know anything about this? (I don't)

odneson Karpathy's MinGPT

These 4 videos (~45 mins) do an excellent job at explaining attention, multi-headed attention, and transformers: https://www.youtube.com/watch?v=yGTUuEx3GkA

odneson Bumblebees’ ‘clever trick’ fools plants into flowering

I looked into it briefly a few months ago but I'm not a biochemist/plant scientist; there is a lot of research on growing human organs, not so much growing fruits.

I'm sure it's possible to trick a plant stem cell into turning into a flower, then fruit, but I think the main question is whether it's economically viable. At the end of the day you would have to bypass photosynthesis and introduce sugars from some external source to grow the fruit which is probably quite expensive and inefficient. I chalked it up to being a very obvious idea which Monsato would have a solution for if it was viable.

Interesting to think though that if you did it perfectly you could turn some weight in sugar + water + stuff into an avocado.

odneson Bumblebees’ ‘clever trick’ fools plants into flowering

I wonder if we'll ever see indoor farms which grow fruits from stem cells, without the need for the rest of the plant.

odneson Can AI Become Conscious?

He talks about this in his appearance on the Lex Fridman podcast. He's my favourite guest, fascinating guy.

odneson A synthetic control causal analysis of Musk's tweet on Tesla stock price

Does anyone know if Twitter are allowed to legally frontrun the markets before making tweets publicly visible?

odneson Avatarify lets users run realtime deepfakes on live video calls

All the new mobile banks seem to do their ID verification/KYC using video selfies. Interesting to consider whether deepfakes are being used in wild to fool these systems and commit fraud.

odneson Agent57: Outperforming the human Atari benchmark

Mind-boggling. They train on each game for a total of 1e11 frames of experience. At 30 FPS that is ~106 calendar years of constant gameplay.

odneson Volkswagen exec admits full self-driving cars 'may never happen'

But Tesla's fleet is getting more driving experience every day than you will get in your entire life, and every time they train on one of those exceptions, the entire fleet will benefit.

Humans can reason about things that haven't to them happened before. Today's machine learning systems cannot. As you say, to react appropriately they must have been trained to do so using human annotated data.

The argument against FSD is that you would need an infinite number of annotated examples, and an infinite number of subroutines for behaving in any identified situations, because the space of driving is effectively infinite.

Until machines are able to do general reasoning about things they've not experienced before then FSD is not happening. By the way, Demis Hassabis thinks that this sort of transfer learning is the key to solving AI.

odneson Volkswagen exec admits full self-driving cars 'may never happen'

I can't see L5 happening without, amongst other things, the human ability to reason about things/concepts we've never encountered before (à la consciousness).

Highway driving is the most constrained normal driving problem, and this is solvable in 99% of cases. But there are so many things that can happen in most other driving situations that make me think that model-based approaches (Tesla) are doomed to fail... Go ahead, train a classifier for every situation you can think of - I guarantee you that you've missed many things.

Elon tweeted the other day that FSD is "coming soon". Either I'm totally wrong about this, and of course I hope I am, or Karpathy + the dev team should be tempering his expectations.

That isn't to say that there isn't immense value in L2/L3, there totally is. But I think that solving driving (being able to drive any situation a human can) is pretty much the same thing as solving intelligence generally.

odneson A Recipe for Training Neural Networks

I was guilty of this once and would add some more specific advice; if your dataset consists of multiple possible labels for the same samples, do not just assume that the average of these labels will describe the best label. And don't assume that training with the unclean data will produce a net that magically learns to do the aggregation for you.

odneson How I Eat for Free in NYC Using Python, Automation, AI, and Instagram

What is the value in unfollowing people? Do accounts that follow a lot of people have less authority?

odneson When Do We Have Empathy for the Mentally Ill?

The opposite of free will is not necessarily pre-determination. It's probably the case that events proceed randomly at atomic levels and this randomness snowballs into more macro events (e.g. neurons discharging).