HNHacker News
TopNewBestAskShowJobs

axg11

3,460 karma · joined May 13, 2020

DMs are open: https://twitter.com/arshamg_
submissionscomments
axg11··on Ask HN: Is it the “worst” time to buy a house?
If I were looking at buying in the next 0-36 months, I would wait at least 6 months. It’s entirely possible that economic conditions continue to slide, interest rates get to 7%+ and a deep recession reduces housing demand. The opposite of all these things could also be true.

If conditions improve faster than expected, you can always quickly enter the housing search/market. Housing isn’t like most other markets that can move 20% in a day. You should be able to wait 6-12 months and closely follow the consensus on interest rates and housing market sentiment.

axg11··on Woman accused of tracking boyfriend with Apple AirTag, killing him with car
To be honest, I was always on the “AirTags are harmless” side until reading your perspective. You are probably right that making the UX so simple and coming from a trusted brand will enable many more people to misuse the technology.

Ultimately, I think it comes down to the question: is this harmful enough that we should regulate it? Answering that involves weighing the benefits against the harms.

AirTags have plenty of legitimate uses and benefits. Stories such as this article show a very real downside to the technology, but isolated stories are not a great way to make an evaluation. Doom, gloom and death will always make a more compelling story than a million people saving 5 minutes finding their keys.

axg11··on Podcast: Housing prices soar with rising inflation
This is essentially Toronto. There is a fairly big delta between mortgage payments (with a reasonable downpayment) and rent. If you only care about monthly cash flow, renting is a far better option. For the last decade the wisdom has been that buying a place is a good idea, despite the cash flow delta, due to appreciation. If people no longer believe this to be true, there could be a sharp shift in the housing market. Perhaps we’re seeing the beginning of that.
axg11··on Tesla: Designing autopilot program to shut off less than a second before impact
What is the alternative? In a situation where the software has detected an imminent and inevitable crash, the _only_ sensible action is to turn off self-driving and revert to 100% human control. I don’t see this action alone as Tesla attempting to avoid culpability.

The only exception to reverting to human control is if the software predicts there is a path to accident avoidance. In that case, the accident-avoiding action should be taken. That’s almost a tautology as it’s a given that FSD software will always _try_ to avoid accidents.

Culpability should be an entirely separate question. What are the expectations of use for a particular FSD software? Tesla is treading a fine line, but my understanding is that you are still expected to pay attention to the road and be ready to take control whenever using FSD.

Perhaps Tesla should also be modelling reaction times? If the time between reverting to human control and the moment of impact is too short, then you could argue the accident is a result of the software. However, what about the time between risk identification and impact time? Different people would identify a risk at different times. The same will be true for different self driving algorithms. FSD has a long tail problem. Very rare events are difficult to learn and test for. The best system we have for dealing with rare events is a human driver - so it’s more than reasonable to me to flip over control to drivers under these circumstances.

axg11··on TBD Web5
Agreed that the current (incorrect) web3 narrative is that all decentralization is good. There will be use cases where it’s good and use cases where it’s bad. The process to a valuable endpoint will involve a lot of duds and failed companies.
axg11··on TBD Web5
It takes a lot to look past the branding, but I think the vision is interesting. I think there _will_ be more personal ownership of data in the future, but not across all services. Decentralized, permissionless self-custody data only makes sense for narrow use cases.
axg11··on New CRISPR-based map ties every human gene to its function
> All genes have a function.

For the benefit of readers unfamiliar with the field: this is wrong.

You can show this experimentally. Construct a gene that produces a non-human protein and introduce this to a human cell/genome. That gene would not have a “function” but still exists in the genome. This is actually happening all the time. Some viruses integrate their genomes when infecting cells. Viral integration is one of the factors that shapes genome evolution.

axg11··on New CRISPR-based map ties every human gene to its function
There’s not any type of mapping and this is what makes biology so difficult to study and understand. One gene can affect anywhere from zero other genes to every single gene. We also need to account for effects at every level of the central dogma of biology. Knockout of a gene can affect genetics, RNA, protein, and much more we don’t even know about yet. The other underestimated complicating factor is that biology is very nonlinear. A 10% increase in gene expression could lead to no resulting effect, whereas an 11% increase in expression could upregulate downstream genes by 1000x.
axg11··on Show HN: Read Wikipedia privately using homomorphic encryption
If someone wants to make false accusations, they can do that now? I can see authorities trying to prevent adoption of homomorphic encryption. Once widely adopted I don’t think merely using it could be used against someone. Similarly, regular encryption isn’t _really_ penalised against now that it’s widely adopted.
axg11··on Show HN: Read Wikipedia privately using homomorphic encryption
Can you explain this comment more? Genuinely asking as I don’t understand the implication/downside?
axg11··on Deploying Transformers on the Apple Neural Engine
Presumably one of:

- the image is never analyzed

- the image is sync'd to device and then analyzed

axg11··on Deploying Transformers on the Apple Neural Engine
I'm only familiar with Android from a distance - does Google Photos perform all the analysis on device?
axg11··on Deploying Transformers on the Apple Neural Engine
Apple is still behind the rest of FAANG in terms of publishing ML research, but they've taken the lead in terms of real world impact of ML. If you go back a couple of years the general consensus was that Apple is very far behind and lacks any credible machine learning groups. As with most of their products, they waited for the technology to mature (a bit) and then made magical experiences from it.

Some examples:

- On-device semantic image search

- Text highlighting (images and videos)

- FaceID

- Computational photography, depth mapping (sensor fusion)

- Background removal (iOS 16)

- Activity detection (Apple Watch)

Siri is still far behind vs. Alexa and Google Assistant, but for every other ML/AI application I'd argue that Apple has the smoothest experience. This should be a lesson for others building ML-powered products. You don't need to use the best, state of the art models to compete. You can compete on the overall experience.

Sharing details and publications doesn't seem to be a core part of Apple's culture, so I'm glad they're going out of their way to publish more details like this article. Their push to optimize ML inference for on-device rather than cloud is going to have the biggest impact on consumer experience. I'm fairly sure this is part of their AR/VR strategy too. Low-latency local machine learning that powers magical experiences.

axg11··on Many researchers were not compliant with their published data sharing statement
This is interesting. Not something that I ever came across but can very much believe this happens.

Most problems in academia come down to bad incentives. Researchers are incentivised to publish for prestige and citations, which are poor proxies for improving our understanding of the world around us.

axg11··on Many researchers were not compliant with their published data sharing statement
As a former (PhD) researcher, this is spot on. Here are the factors I've observed:

- fear of getting scooped: there's a general fear that sharing data makes it easier for competing labs to publish on the same topics, leading to the publishing lab "missing out" on publications

- fear of being wrong: sharing the data leads to more scrutiny, leading to a higher chance of a fatal flaw being found

- extra work: publishing data often involves jumping through extra hoops with little to no personal gain

axg11··on A six-month program in Britain will assess the effects of a shorter workweek
What if, at a societal level, we just end up employing more people on fewer days? Wouldn’t that be a good thing?
axg11··on Toronto, the Quietly Booming Tech Town
Toronto is a no-brainer for most US tech companies. Cheaper to hire, abundant talent, similar (enough) culture, accessible time zone and very permissive immigration laws if you want to bring in more talent.

Salaries are increasing as a result, although the benefits have been masked by inflation, especially for real estate. The long term question is whether the increased demand for Toronto tech talent from US companies translates into more Canadian startups and successful Canadian companies.

axg11··on Norway to track all supermarket purchases?
High trust society, as in: there is a high degree of trust between any two random people that they are likely to act in each other’s benefit when it comes to societal issues. That is fundamentally how I view taxes. We’re contributing to a common fund for common good.
axg11··on The economics of stadium names
Crypto.com 24h trading volume is ~$400m, so they're doing $175B in trading volume every ~450 days. By those metrics, the sponsorship deal doesn't seem unreasonable? Clearly they're not getting $400m in incremental volume as a result of the sponsorship but over 20 years the ROI is in the right magnitude. It's also notable that retail traders (the target audience of a sports sponsorship) are the most lucrative and pay the highest fees.
axg11··on Norway to track all supermarket purchases?
There's a general reaction that big government is bad, more tracking is bad. Norway ranks highly in most metrics of development and quality of life. Perhaps in a high trust society this type of data collection is a positive.
axg11··on Norway to track all supermarket purchases?
The people in society who need help the most often have the least trust in government and don't have the means to understand and navigate programmes that could help them.
axg11··on Former OpenSea exec arrested and charged with insider trading of NFTs
I'm not a lawyer. From a little research I understand that insider trading laws apply to securities, so I can only conclude that NFTs are being defined as securities here. It's always been a grey area.
axg11··on FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
I agree they're probably here to stay. It's interesting to think about how the hardware-software feedback loop affects architecture choice and research directions. I'm almost certain there is an even more impressive architecture out there, waiting to be discovered. However, until we maximize the performance we get out of Transformers, there is a strong incentive to optimize both software and hardware specialized to this architecture.

One downside is that as hardware/software becomes more specialized to one architecture, it becomes more difficult to explore very different architectures.

axg11··on No, DALL-E doesn’t have a secret language
Exactly this. At a high level, DALL-E is mapping text to a (continuous) matrix and then mapping that matrix to an image (another a matrix). All text inputs will map to _something_. DALL-E doesn't care if that mapping makes sense, it has been trained to produce high-quality outputs, not to ensure the validity of mappings.

None of this makes DALL-E any less impressive to me. High quality image generation is a truly amazing result. Results from foundational models (GPT-3, PaLM, DALL-E, etc) are so impressive that they're forcing us to reconsider the nature of intelligence and raise the bar. That's a sign of a job well done to me.

axg11··on Ask HN: Are companies preparing to end Remote?
Companies that advertise roles on HN are not representative of the economy.
axg11··on Elon Musk on Remote Work
Whichever side of the remote vs. in-person debate you fall on, I think this is a net positive for the industry. More companies are starting to become strongly opinionated on the topic and that leads to a wider spectrum of options when searching for a job. There are many people out there who hate fully-remote and are suffering as a result of the isolation. There are also people out there who are thriving due to remote work. Both can co-exist at different companies.

Of course, springing this on already remote workers is problematic. Tesla is no stranger to low employee retention. In the short term they will have an exodus but perhaps in the longer term they'll have a workforce more aligned to their culture.

axg11··on Google's PageRank patent has expired (2019)
In general the biggest breakthrough in that space has been machine learning powered relevance scoring and NLP. PageRank worked by using high level metadata (links between pages). We now have methods that can incorporate metadata information and analysis of the content itself.
axg11··on DALL-E 2 has a secret language
Indeed, this is an expected outcome from the way that DALL-E is trained. It’s still an interesting finding and sometimes affirming the expected can lay the groundwork for future interesting discoveries.

This is an example of an application where uncertainty modelling would help greatly. Any and every input will lead to an output. That doesn’t mean that all regions of latent/embedding space are equally valid.

I’m in the camp that large/modern ML models are nearing human intelligence, in some aspects. What’s currently missing is the universal ability to estimate uncertainty and identify inputs that are out of distribution. Many groups are working on this and perhaps we already have the solution but are not combining the right uncertainty estimation approach with the right foundational model.

axg11··on A PaaS dedicated to indie hackers and small teams
FYI the site is quite broken on mobile (iOS) - it’s difficult to read.
axg11··on Workers quitting over return-to-office policies
I like the concept of async work but I’m also scared of how transactional that would make work. Most companies that preach async now have a very strong, “friendly” culture. If async catches on more widely I’m worried it would be the final push to make software development purely transactional and performance/pay would be directly tied to metrics.

I enjoy the human component of work. I don’t like _all_ of my coworkers, but it’s still a net positive to interact with them.

When I think about the ideal software job, a lot of the aspects of day-to-day work informally become async. Especially as you work with people for longer periods of time and gain mutual trust, a good workplace allows both managers and individual contributors to structure their own time.

← PreviousPage 5 of 17Next →