HNHacker News
TopNewBestAskShowJobs

axg11

3,460 karma · joined May 13, 2020

DMs are open: https://twitter.com/arshamg_
submissionscomments
axg11··on Ask HN: If Web3 just a P2P database is it incapable of fulfilling its promise?
Web3 is a loosely defined trend and every person has different expectations for the changes that web3 should bring. Here's what I personally think web3 can do to change the internet for the better:

- Increase proportion of consumer spending that goes to creators (Apple, YouTube, Spotify, etc have egregious take rates from 20-50%)

- Create new business models where incentives are aligned. Take a look on HN at the weekly threads about poor Google search results. Subscriptions and micropayments are a no-go for services like Google search. It's time for a new business model for these services and web3 provides the framework to experiment with radically different business models.

- Permissionless (and decentralized) finance: if you live in the US/Canada/UK/France/Germany/etc then this might not be much use to you. There are dozens of countries that can benefit from permissionless access to banking.

axg11··on I spent two years launching tiny projects
There are a few SaaS templates out there and services such as Memberstack that make it easier to implement the boilerplate SaaS features in Webflow. I'd also recommend just doing the boring work once and keeping that as a template for future projects.
axg11··on U.S. court rejects Tether’s bid to conceal reserve records from the public
USD collateralized stable-coins are happening whether central banks participate or not. I think the advantage for a central bank would be to bring stability and increase demand for the underlying asset (the centrally controlled currency).
axg11··on U.S. court rejects Tether’s bid to conceal reserve records from the public
I agree that 1% (if true) is not insignificant, but destroying all cryptocurrencies to reduce our energy consumption by 1% isn't a clear tradeoff to me. We also need to account for the fact that renewable energy as a percentage of energy used by crypto is increasing and already at significant proportions.
axg11··on U.S. court rejects Tether’s bid to conceal reserve records from the public
Aside from optics, why doesn't a major central bank step in and create a stable-coin? A GBP/CAD/USD stable-coin collateralized by a central bank's reserves would be widely adopted at this point.
axg11··on U.S. court rejects Tether’s bid to conceal reserve records from the public
A great, laconic statement to rally the anti-crypto troops. In reality, destroying cryptocurrencies would have zero effect on any of those three:

Estimated total energy usage for all cryptocurrencies: 0.1-1% of global energy usage

Inflation: I don't see a link whatsoever - any influence on inflation from crypto is dwarfed by a decade of quantitative easing.

Labor supply: estimated number of developers working in crypto is <100,000, whereas estimates for total number of software developers in the world is ~20M, so ~0.5%.

axg11··on AI recognition of patient race in medical imaging: a modelling study
Race _is_ a spectrum but genetic differences themselves are distinct (SNPs). It's trivial to train a classifier to distinguish race from genetic data, hence, I'd argue they are distinct groups.

You can draw an analogy to colours in the rainbow: a rainbow is a spectrum but we can still draw lines that demarcate colours. Colour definitions are fuzzy at the edges but this doesn't mean coarse colour labels are not distinct.

axg11··on Gato – A Generalist Agent
You are describing most successful machine learning models. Take AlphaFold, it has surely discovered relationships that govern protein folding better than any human has ever previously understood.
axg11··on AI recognition of patient race in medical imaging: a modelling study
Perfect example of citations-driven research. The authors aren’t motivated by a genuinely interesting scientific question (“are anatomical differences between genetically distinct groups of people visible in X-rays?”). Instead, the authors know that training a classifier to predict race will generate controversial headlines and tweets. All publicity, positive or negative, leads to more citations.
axg11··on Riot Games: Artificial Latency for Remote Competitors
I used to work in HFT. I promise you that companies would still try to exploit randomized batches. There is an advantage to being the very last entrant into a batch (most up to date information). Truly random batches are not trivial to implement and any statistical pattern in the batching could be exploited.
axg11··on AI recognition of patient race in medical imaging: a modelling study
> ML models are great tools, but they're way too much of a black box.

A human doctor is also a black box, in meat form.

axg11··on Riot Games: Artificial Latency for Remote Competitors
That sounds like a complex solution. Sometimes a dumb solution that works good enough is better than a complex solution that _probably_ can't be exploited.
axg11··on OpenAI Codex Python to C++ Code Generator
A question for the computer scientists here: are there applications for this where Codex could be used to automatically optimize complex systems?
axg11··on Teaching GPT-3 to reverse words
100% agreed. I already see myself doing this with Github Copilot. If I write a comment or start a line of code in a certain way, I get a much better suggested code completion.
axg11··on Tesla FSD Beta almost causes a head-on collision [video]
> "That was my first near-accident on beta"

As a tech and machine learning maximalist I'm frustrated with how haphazardly Tesla is approaching self driving. This is not a technology that should be QA'd consumer-first. On top of that their approach of ditching LIDAR and other sensors is shooting themselves in the foot.

Tesla is in danger of putting the entire AV industry back by ~5 years just through loss of reputation. Regulators and AV pessimists will jump on the opportunity to use Tesla as an example of why these vehicles should be completely banned from public roads. Cruise, Waymo, Zoox, et al. are testing and rolling out responsibly while Tesla is using their customer base as crash test dummies.

axg11··on Web3 is expensive P2P
I agree - misguided attempts to make everything "web3" shouldn't take away from the huge amount of value that future web3 services can bring, even if they are vastly outnumbered by web2 services forever.
axg11··on Web3 is expensive P2P
> But successful internet communities do everything in their power to remain walled gardens. What could they possibly have to gain by doing this? Like what would Reddit gain by moving Reddit Gold to the blockchain?

We have to reimagine everything, I don't think a successful web3 platform will emerge from simply tacking on tokens to existing platforms. Successful internet communities become walled gardens to protect against competition, but they're limiting the growth of their community members and I believe the platforms themselves could capture more value. A creator on TikTok today can't easily give their most loyal followers discounts on their Shopify store, perks for moderating a Discord, early access to YouTube content, etc.

axg11··on Web3 is expensive P2P
I'm not sure I understand your question, genuinely. There are of course costs for the benefits that come with web3: protocols are slow to evolve, and infrastructure is less efficient when we require a distributed consensus. Anyone that thinks "Web 2" or centralized systems will disappear is silly, there will always be situations where the costs of web3 are greater than the benefits. On the other hand, it's equally silly to think that there are zero use cases where the benefits of web3 outweigh the costs. Both can and will coexist.

Static HTML pages still exist today.

axg11··on Web3 is expensive P2P
It's strange that people lump these four together. There's a marked difference between Facebook/Instagram (friends first, sharing encouraged by all users) vs. Twitter/TikTok (mainly acting as a broadcast medium with a long tail).
axg11··on Web3 is expensive P2P
Some benefits that I think web3 can bring to the user:

- Universal login: An alternative to "Sign in with Google/Apple/XYZ" that can't be denied by a corporation. There are countless examples of these companies deactivating accounts for questionable reasons.

- Permissionless composability: I'm not sure how to word this one better. If a developer wants to build on top of blockchain data, they can. No need to apply for API access, rate limits, etc. Not all data should be open/on a blockchain, but in the cases where it's appropriate this is a fantastic benefit.

- Ownership in communities: tokens (fungible or non-fungible) can represent ownership for communities. The next Reddit will reward its community and contributors. This isn't an unusual concept. YouTube and TikTok already reward their creator communities. The difference is that existing platforms don't let you build on those communities outside of their walled garden.

For balance, here are two concepts/terms I see co-mentioned with web3 that I believe are bullshit:

- Metaverse: I don't think this will come to anything. We're already living in the metaverse. We have online identities (yours is mouzogu!). I refuse to spend a significant portion of my life in VR or another 3D game.

- Micropayments: Consumers don't want the mental overhead of figuring out value for money every time they consume content.

axg11··on Modern JavaScript Tutorial
Is there a list of equivalent excellent resources for other languages? I'm familiar with JS and Python - I'd love to get up to speed with other languages this way.
axg11··on The Tabula Sapiens: A multiple-organ, single-cell transcriptomic atlas of humans
If you’re interested in questions about effective science funding, single cell transcriptomics is an interesting field to study.

Quick primer for those unfamiliar with single cell transcriptomics (ELI18 level): it’s a technique where you take one cell at a time and measure which genes are on (expressed) and measure the level of expression. The resulting data is _very_ high dimensional. Typical experiments these days result in millions of cells and tens of thousands of gene expression values (after much processing).

Since single cell RNA-seq (scRNA-seq) emerged there’s been a focus to “map” the transcriptome of every possible cell type, in a quest to understand more about different cell types and the relationships between them. Huge amounts of grant money has poured in from large science funders, notably including the Chan Zuckerberg Initiative. The technology has improved rapidly over the last decade, and it’s now common to see scRNA-seq experiments tacked onto papers/studies that wouldn’t have such data 3+ years ago.

There is a negative side to the rise of scRNA-seq data however. Since the data is inherently high-dimensional, it can be difficult to interpret. Every single cell transcriptomic experiment produces results, regardless of whether there is interesting underlying biology to be revealed. A common analysis flow in the field is to apply dimensionality reduction (PCA, tSNE, UMAP) to the data and then interpret the resulting clusters/mapping. This is really problematic as it’s trivial to show that random data can produce clusters. I don’t want to paint the picture that all groups analyse scRNA-seq incorrectly. However, the rise in availability of the technology far outpaced the field’s ability to analyse the data.

These issues don’t apply to the Human Cell Atlas project, which has specialists analysing the data and much more internal/external scrutiny on the data. However, there’s still a question of lacklustre scientific output. Large projects such as the Human Cell Atlas are perfect for advancing scientists careers. They are too big to fail. Owing to the difficulty in interpreting scRNA-seq and the mapping/atlas goal of the project, there is no way for it to fail. Any output data is taken as gospel.

I don’t want to be too cynical. Mapping all human cell types is a genuinely interesting scientific question. On the other hand, I’m not sure that the way these mega-projects are structured produces the best science. On the optimistic side, I think mega projects do significantly help technology development. The Human Cell Atlas created huge demand for scRNA-seq technologies and products, much in the same way that the Human Genome Project did before it. I think in the long term that will be the lasting positive impact, rather than the results themselves.

Context: I have a PhD in single cell transcriptomics from a lab that went on to get a large grant as part of the Human Cell Atlas project.

axg11··on Tech bubbles are bursting all over the place
Do you know which website you’re on?
axg11··on DeepMind: A Generalist Agent
I think AI will largely remain an input-output tool. We still need to prepare ourselves for the scenario where for most input-output tasks, AI will be preferable to humans. Science is an interesting field to focus on. There is so much scientific literature for most fields that it is now impossible to keep up with the latest literature. AI will be able to parse literature and generate hypotheses at a much greater scale than any human or team of humans.
axg11··on DeepMind: A Generalist Agent
You're right. I didn't word that very well. Human intelligence vs. AI will always have different qualities as long as one is biological vs. silicon based. I still think we'll be surprised how quickly AI can catches up to human performance on most tasks that comprise modern jobs.
axg11··on DeepMind: A Generalist Agent
Slowly but surely we're moving towards general AI. There is a marked split across general society and even ML/AI specialists between those who think that we can achieve AGI using current methods and those who dismiss the possibility. This has always been the case, but what is remarkable about today's environment is that researchers keep making progress contrary to the doubter's predictions. Each time this happens, the AGI pessimists raise the bar (a little) for what constitutes AGI.

Just in the last five years, here are some categories of pessimistic predictions that have been falsified:

- "AI/ML can't solve scientifically useful problems" - then AlphaFold changed the protein folding field

- "We're entering an AI winter" [0] - then transformers continued to show promise across multiple domains

- "ML models can't perform creative work" - then came GANs, large language models, DALL-E, and more.

- "Generative ML models are just memorizing the dataset!" - then came multiple studies showing this to be false for well trained GANs, diffusion models and other types of generative models. Take a look at DALL-E 2 generated images of "a bear putting on a shirt in H&M".

- "AGI is impossible - look at language models, they have no understanding of the world and make silly mistakes" - the second part is true, large language models are artificially limited due to being language-focused. Nowadays there are approaches such as Gato and other multi-modal models. Humans develop intuition through multiple sources of information: sight, sound, smell, and touch. Given enough multi-modal context I'm confident multi-modal models will be able to show human-like intuition.

I'm not anti-skeptic. Skepticism is essential to all good science. I think the danger of skepticism with respect to AGI is that we're being complacent. Given the trajectory of improvements in machine learning, we should start preparing for a world where AI is indistinguishable, or far superior, to human intelligence.

[0] - https://www.bbc.com/news/technology-51064369

axg11··on Airbnb May 2022 Release
I don't understand what you mean by unusable. I just went to Airbnb.ca, typed in a postal code in the very big entry form at the top, chose the next two nights and immediately received relevant results.
axg11··on Coinbase Q1/22 Shareholder Letter [pdf]
Layoffs? Maybe. Paycut? Not at their existing companies. There's a case to be made that too many tech companies scaled too quickly. Fundamentally, Coinbase is still a solid company, they may have overextended hiring but their core businesses are very profitable.
axg11··on Coinbase Q1/22 Shareholder Letter [pdf]
They are trying to do everything:

- Professional trading platform (Coinbase Pro)

- On-ramp for the masses (Coinbase)

- NFT marketplace (Coinbase NFT)

- web3 infrastructure (Coinbase Cloud)

- Debit card w/ crypto rewards (Coinbase Card)

- Crypto for businesses (Coinbase Institutional / Coinbase Custody)

Each of these is a huge undertaking, often across multiple different markets across the globe.

axg11··on The day I discovered that Apple Maps is Kind of Good now
Apple Maps becoming good is a testament to Apple’s engineering culture and relentlessness. They made the decision, 9 years ago, to provide an alternative to Google Maps. They kept iterating on the product despite being inferior during the earlier days. Most other big tech co’s printing the amount of cash that Apple has would have abandoned the maps product and paid the Google tax.
← PreviousPage 7 of 17Next →