HNHacker News
TopNewBestAskShowJobs

axg11

3,460 karma · joined May 13, 2020

DMs are open: https://twitter.com/arshamg_
submissionscomments
axg11··on Web3 is the big idea customers didn’t ask for?
Two novel ways:

1) they're funding previously biotech companies (e.g. traditional venture not interested because insurance companies won't reimburse hair loss treatments)

2) incentivizing scientists (and other workers) from anywhere in the world to work part time towards these scientific goals

This is only possible due to the token incentive schemes and DAO governance structure. Crypto was never envisaged as a way to improve science, but it's being adopted anyway.

axg11··on Web3 is the big idea customers didn’t ask for?
You're looking in the wrong places. Look up VitaDAO, Molecule, HairDAO - three projects making real scientific impact enabled by crypto.
axg11··on Web3 is the big idea customers didn’t ask for?
This is a much more level-headed take on web3. As someone building in web3, I admit that 99% of current web3 projects (including tokens and NFTs) will go to zero. That's a consequence of the sheer number of projects and diversity of experimentation occurring today. For the first time in a decade, there is an explosion in the number of products and projects being developed. I can only see this as a positive thing.
axg11··on Show HN: I ranked news websites by speed
I didn’t intend to use engagement metrics in a negative way. We also have to note that the audience has changed. The audience of the early web was very different to the average audience today (almost everyone).
axg11··on US military's mysterious X-37B space plane zooms toward orbital record
The best publicly available information just isn’t very good. We can speculate on the purpose of the X-37B programme but we won’t know for sure, perhaps for decades.
axg11··on US military's mysterious X-37B space plane zooms toward orbital record
It’s not a meme, it’s simply following history and reading declassified materials. The US government has a track record of developing advanced aircraft and weaponry in secret.
axg11··on Show HN: I ranked news websites by speed
You're not wrong. This is how the internet used to look like. Modern web layouts and designs evolved to be this way because they result in higher engagement metrics.
axg11··on Human-specific gene increases primate neocortex in the fetal marmoset
Biotech/genomics and AI/ML are following very interesting parallel trajectories over the last ~10 years.

In biotech, we have developed incredible tools: DNA editing (CRISPR), synthesis, and sequencing. We have a lot of tools but we haven't cracked systems level understanding yet.

In AI/ML, we have developed algorithms that can ace tasks thought impossible just 15 years ago. We have a hunch of how and why they work, but we lack deep understanding.

axg11··on Meta is inviting researchers to pick apart the flaws in its version of GPT-3
It really is great that Meta released the notes/logbook. Credit where credit is due. Very few other academic or industry labs release materials like this, especially when the reality is so messy.

Some interesting takeaways:

- Meta aren't using any software for scientific logbooks, just prepending a document

- So many hardware/cluster issues.

- Hot-swapping algorithms is common and likely underreported (in this case activation functions and optimization method)

- A well resourced team didn't solve enough issues to fully utilize compute resources until >50% of the total time into the project

axg11··on Who is going to replace Google for us?
Having worked for a different FAANG, I saw the immense pressure to incorporate data from other products. I imagine there’s a big internal push to aggregate all data from across Google services. It’s an easy win for product managers looking for projects.
axg11··on What Copilot means for open source
I believe this is already a feature that you can activate for Github Copilot.
axg11··on Sentence Embeddings have a problem, the reason sometimes Dall-E2 fails
Is using a single embedding sufficient to capture the entire meaning of a sentence?

High similarity scores between sentences is not the whole picture. A sentence and it’s negative/opposite are naturally very similar in topic and sentence structure. The mapping from text embedding to image needs to be aware of the most important dimensions of the embedding when constructing the output image.

axg11··on Returning to the office is costing you untapped talent
I don’t think high unemployment is a certain outcome. Layoffs are trending higher, but not because of bad economic metrics. Companies are laying off employees in anticipation of bad times. There’s a scenario where inflation returns to lower levels and the economy doesn’t deteriorate.
axg11··on YaLM-100B: Pretrained language model with 100B parameters
> We're a long, long way from this.

We’re probably 18 months away from this. We’re probably less than 5 years away from being able to do this on local hardware. AI/ML is advancing faster than most people realise.

axg11··on Tech talent is still in demand but outsized salaries may be disappearing
Don’t be worried at all. Software has been one of the most in demand skills for the last 10 years and will still be over the next 10 years. Salaries at the very top end might go down. Open vacancies might reduce. Even in the worst case scenario, any software engineer will be in a much better position than 95% of other professions. Economic recessions affect many different industries.
axg11··on Ask HN: Why is Lex Fridman so successful with his “tech” podcasts?
Listen to his interview with Mark Zuckerberg. That interview demonstrates why he is really special and nobody else is doing interviews like him:

1. Who else is landing guests like Zuckerberg?

2. Of those people, who has the willingness to ask questions such as "why do you think you are one of the most unpopular CEOs on Earth?" (paraphrasing)

3. Of the interviews that fit (1) and (2), how many are holding conversations free of a personal agenda? Lex's interviews are generally judgement-free. He's more interested in exploring deep questions than presenting one single political view.

So, even though he can come off as boring or monotonic at times, he's rather unique as an interviewer. That is what makes him successful.

axg11··on SpaceX said to fire employees involved in letter rebuking Elon Musk
Most employees never consider starting their own company, so we can strike off (2). Buying SpaceX is vanishingly unrealistic, so discount (3). My impression of working at SpaceX is that the options for employees are:

1. Keep working at SpaceX - work on truly innovative technology with a lofty mission. The work environment sucks, you work 100 hour weeks but the _work_ itself is great. Your work might land on Mars this decade.

2. Join another startup (Relativity, Firefly, Rocket Lab, etc) - no proven track record of success or work on smaller scale (but successful) projects. Work hours and culture are variable, but there is a general sense of urgency. Your work is not landing on Mars this decade but could still change the aerospace industry in smaller ways.

3. Coast and enjoy life with your family (Lockheed Martin, Boeing, Blue Origin, etc). Work 38 hour weeks. You will get the chance to work on large prestigious projects. Your project is regularly in the news for being over budget and late. There is no sense of urgency. You have complete job security.

axg11··on SpaceX said to fire employees involved in letter rebuking Elon Musk
This is a very predictable consequence of criticizing your employer via a public letter. Criticize internally all you want, and influence the changes that you want to see happen. Employment is a two-way relationship. If you don't like your employer, you are free to leave. If they don't like you, they are free to fire you (within legal bounds). Publishing an openly critical letter and signing it is a quick way to get your employer to not like you.
axg11··on Why am I no longer qualified to be a Brex customer?
Why does the cost have to be competitive? Price it at a level that covers all costs. If the price is uncompetitive, 99% of users will not take it, however, that 1% that _really_ need to continue using Brex (for whatever tail reason) can continue to until they're ready to migrate.
axg11··on Why am I no longer qualified to be a Brex customer?
Why do this rather than offer customers a chance to continue service at some higher cost?
axg11··on LaMDA is not sentient
I'm really happy that large language models are forcing us to reconsider these questions. I don't think we can be sure of the sentience of anything other than ourselves, almost by definition. Having said that, I go about my life _assuming_ that the other humans I encounter are sentient.

At what point will we have enough evidence to _assume_ that AI is sentient?

P.S. I am not arguing that LaMDA is sentient. I don't think anyone credible or serious is arguing that LaMDA is sentient.

axg11··on Ask HN: Why does WWDC get 10x more views than Google I/O?
Aren’t you just proving my point? For each of these categories, Apple’s entry massively expanded the entire category and redefined the form of all future post-Apple products.
axg11··on Ask HN: Why does WWDC get 10x more views than Google I/O?
You’re correct - Maps is a good example. I think the relative point still stands though. On the whole, Apple releases fewer higher-signal products and a better track record of industry wide impact.
axg11··on Ask HN: Why does WWDC get 10x more views than Google I/O?
Apple has a history of surprises and engaging keynotes. Google, for all its success, has never launched a category creating product. Every Google success I can think of is a follow-on product, even search.

Apple’s category creation resume:

- iPod

- iPhone

- iPad

- Apple Watch

- AirPods

- ARM Mac/Apple Silicon (still early but looks like an industry shift)

Apple’s successes above weren’t necessarily the first in the category, but they were the first to make the category significant. Watching the corresponding Apple keynotes for each of these launches was a window into the future. What’s the advantage of watching Google IO when they’ve rarely if ever dictated the future?

axg11··on Neural networks 'learn' the way humans do? A neuroscientist explains why not
We don’t even know what sentience is. We don’t have a good way of defining sentience, nor a common definition. If we were to definitively create sentience, there wouldn’t be a universal consensus on whether we had actually created it.
axg11··on Neural networks 'learn' the way humans do? A neuroscientist explains why not
Yes, thank you for explaining this more clearly :)
axg11··on Neural networks 'learn' the way humans do? A neuroscientist explains why not
The focus on learning is misguided. Human brains don’t have to use the same mechanism for learning as they do for reasoning. Perhaps they do, but I don’t believe we have enough evidence either way.

The useful part of AI is the inference/output. If we achieve human-level performance AI inference then we have an incredibly powerful tool that can revolutionise society. The learning algorithm, be it gradient descent, evolutionary mechanisms or something entirely new, doesn’t matter. Whatever gets the job done.

axg11··on Stuxnet is embarrassing, not amazing (2011)
Long term effects of Stuxnet are unclear. In the short term, it was a huge win in delaying the programme and sending a message that even air gapped systems are at risk. In the long term, Stuxnet put the programme and every adversary on notice that they should increase focus on security practices.
axg11··on Stuxnet is embarrassing, not amazing (2011)
How could it remain hidden? The programme heads would think it was an accident? This was a devastating setback to the entire programme. There would surely be an investigation to ensure a repeat could not happen.
axg11··on Stuxnet is embarrassing, not amazing (2011)
Hiding and obfuscating the payload is pointless in this case. The author doesn’t seem to understand the reasons why a group would go through the trouble of obfuscation or other similar techniques.

Stuxnet was a one-time operation with a very limited opportunity window. Target systems were airgapped. A large part of the success of the operation relied on a human penetrating that air gap. A successful operation would be attributed to either Israel or USA immediately. What is the benefit of obfuscation?

← PreviousPage 4 of 17Next →