HNHacker News
TopNewBestAskShowJobs

PheonixPharts

4,161 karma · joined April 7, 2022

submissionscomments
PheonixPharts··on Should I use a pie chart?
Also cases where someone wants to visually represent exactly how much of a pie they've eaten.
PheonixPharts··on Should I use a pie chart?
Can you give me a single example where a pie chart would better convey information than a table?

The only visualization more useless than a pie-chart is a 3d pie-chart.

PheonixPharts··on Should I use a pie chart?
> I had to google to find some arguments against pie charts

Can you give me any arguments for pie charts? I also disagree with that article that the best alternative to a pie chart is a bar chart.

Whenever your doing any sort of data visualization you should always ask "How does this representation convey information that cannot be represented in a table?"

Can you provide an example where a pie chart would tell you more than could be conveyed in a table?

Tables are incredibly information dense, have an unambiguous ordering and can easily scale to expressing 5-6 dimensions of information in the exact same space that a pie-chart can convey. Tables are limited in that they work best when representing a very small number of observations, but pie-charts and bar-charts share this same problem without any of the benefits of tables.

If a table won't cut it, then you probably have enough data that a scatter plot is your best choice. Scatter plots also can easily handle an extra 2-3 dimensions (color, scale, and shape) of information.

PheonixPharts··on SQL for data scientists in 100 queries
You must have worked at different places from me. Nearly every DS job I had (before wisely apparently) leaving that area had leetcode style algo questions during the interviews.

Again, things have apparently changed.

PheonixPharts··on SQL for data scientists in 100 queries
Not to detract from the article, but: Wow the meaning of the term "data scientist" has changed since the days of "sexiest job". From the article description:

- Rachel has a master’s degree in cell biology and now works in a research hospital doing cell assays.

- She learned a bit of R in an undergrad biostatistics course and has been through the Carpentries lesson on the Unix shell.

- Rachel is thinking about becoming a data scientist and would like to understand how data is stored and managed.

Data Scientists, back in the day, were largely people with both a fairly strong quantitative background and a strong software engineering background. The kind of people who could build a demo LSTM in an afternoon. Usually there was a bit of a trade-off between the quant/software aspects (really mathly people might be worse coders, really strong coders might need to freshen up on a few areas of mathematics), but generally they were fairly strong in each area.

In many orgs it's been reduced to "over paid data analysts" but I wouldn't even hire "Rachel" for a role like that.

PheonixPharts··on Markov Chains are the Original Language Models
LLMs are Markov chains in latent space, it's the latent representation that give them their power, but ultimately there's not as much difference as one would suspect.
PheonixPharts··on I don't want anything your AI generates
I personally think AI will be it's most powerful not as some generated output, but as invisible glue that binds parts of a larger system.

AI (specifically the LLM variety) should be performing small tasks via agents and then using structured output to allow those agents to pass information in a larger system. There, as an example, countless zero/few shot classification tasks that LLMs crush traditional ML at. You want user tickets routed to the correct rep? That sounds like a task an LLM should be doing behind the scenes.

Code gen as an output is likewise boring, agents that adaptably learning to code themselves, generate tests, allow for debugging etc, that has the potential to be very powerful.

Unfortunately I still feel the next AI winter will hit before people even really scratch the surface of how these tools can be used in practice. Everyone is tragically trapped in the prompt -> output model, rather than really thinking about agent based approaches.

PheonixPharts··on Non-deterministic execution of Python functions
Makes me think about how non-deterministic computing is basically how you view lists as a context (as opposed to a container) when thinking in terms of Monads in Haskell. Application functors basically form a minimal framework for applying functions to values in a non-deterministic context when working with lists.

Making this further relevant is that perhaps the most well known example of monadic programming of lists is: Python's list comprehensions. Which makes a bit surprised to see that not more explored/exploited in this library.

Interesting nonetheless!

PheonixPharts··on Meta AI releases Code Llama 70B
I imagine their goal is to simultaneously show that Meta is still SotA when it comes to AI and at the same time feed a community of people who will work for free to essentially undermine OpenAI's competitive advantage and make life worse for Google since at the very least LLMs tend to be a better search engine for most topics.

There's far more risk if Meta were to try to directly compete with OpenAI and Microsoft on this. They'd have to manage the infra, work to acquire customers, etc, etc on top of building these massive models. If it's not a space they really want to be in, it's a space they can easily disrupt.

Meta's late game realization was that Google owned the web via search and Apple took over a lot of the mobile space with their walled garden. I suspect Meta's view now is that it's much easier to just prevent something like this from happening with AI early on.

PheonixPharts··on I used to not worry about climate change. Now I do [video]
> "nothing matters the world is doomed soon"

What you're describing is people experiencing grief. I'm what you would call an "extreme doomer", and did feel this way about a decade ago when I first realized the dire situation we are in.

But your first reaction is not the end. Waking up and accepting our state, and the despair that came with it, was the first stage in a long process of learning to live, and be comfortable in the world the way it is.

What you are suggesting is that we just ignore grief and pretend nothing is happening. I suppose this is just the denial stage of grieving, but it's worth recognizing that the people you know feeling this way are working through a process.

Pretending that someone isn't dead isn't a long term solution to accepting loss, even if it means in the short term you are in a dark place emotionally. It's unfortunately it upsets you to see people experiencing this, but those people, given enough time, might be in a better place to help you when you finally have to let go of denial.

PheonixPharts··on I used to not worry about climate change. Now I do [video]
The paleontologist Peter Ward has made a fairly compelling argument (in both academic research and the popular science book Under a Green Sky), that virtually all major extinction events are related to sudden rises in atmospheric CO2. A lot more happens when CO2 rises quickly other than it getting warmer. A major issue is it's impact on the oceans, which can become anoxic [0], causing them to emit hydrogen sulfide instead of oxygen. Such an event would be devastating to our current ecosystem.

> There won't be an extinction event.

This also ignores the fact that we are currently in the 6th largest extinction event in the history of life on this planet [1]. Elizabeth Kolbert's The Sixth Extinction is a great book on this (and the history of our understanding of species extinction as well).

So aside for being naive about the science, your comment reads a bit like claiming you don't think it will rain today while in the midst of being soaked in a massive rain storm.

0. https://en.wikipedia.org/wiki/Anoxic_event

1. https://en.wikipedia.org/wiki/Holocene_extinction

PheonixPharts··on Implementing a ChatGPT-like LLM from scratch, step by step
I honestly cannot fathom why anyone working in the AI space would find $50 too much to spend to gain a deeper insight into the subject. Creating educational materials requires an insane amount of work, and I can promise, no matter how successful this book is, if rasbt were do the math on income generated over hours spent creating it wouldn't make sense as an hourly rate.

Plenty of other people have this understanding of these topics, and you know what they chose to do with that knowledge? Keep it to themselves and go work at OpenAI to make far more money keeping that knowledge private.

If you want to live in a world where this knowledge is open, at the very least refrain from publicly complaining about a book that cost roughly the same as a decent dinner.

PheonixPharts··on Implementing a ChatGPT-like LLM from scratch, step by step
Automatic differentiation is why we are able to have complex models like transformers, it's arguably the key reason (in addition to large amounts of data and massive compute resources) that we have the revolution in AI that we have.

Nobody working in this space is hand calculating derivatives for these models. Thinking in terms of differentiable programming is a given and I think certainly counts as "from scratch" in this case.

Any time I see someone post a comment like this, I suspect the don't really understand what's happening under the hood or how contemporary machine learning works.

PheonixPharts··on Implementing a ChatGPT-like LLM from scratch, step by step
Reinforcement learning is an entirely separate area of research from LLMs and, while often seen as part of ML (Tom Mitchell's classic Machine Learning has a great section on Q learning, even if it feels a bit dated in other areas) it has little to do with contemporary ML work. Even with things like AlphaGo, what you find is basically work in using deep neural networks as an input into classic RL techniques.

Sutton and Barto's Reinforcement Learning: An Introduction is widely considered a the definitive intro to the topic.

PheonixPharts··on The rise of batteries in six charts
It's not "exponential" it's "exponential in the S-curve", which is just some bizarro marketing way of saying "sigmoid growth".

Sigmoids (the most well know being the logistic curve) begin to tapper off overtime approaching no growth and reaching an upper bound.

PheonixPharts··on An open source DuckDB text to SQL LLM
> but it's actually kind of hard to get working in practice

One of the biggest challenges I've personally seen in this space is business "leaders" pushing teams to ship products asap lest they loose face among their fellow CEOs for not pushing out "AI" products before everyone else.

I'm fairly optimistic about LLMs being able to truly be transformative, but it's not going to be through forcing the bread-dead UX of hoisting yet another slightly re-imagined chat interface on users.

The idea of "talking to your data" is a promising one, and anyone who has worked for a large data driven org will quickly agree that organizing and searching in-house data is not a solved problem from the UX end of things. But to truly solving these problems, even/especially with LLMs, is going to require thought and experimentation. Something few "business leaders" have patience for.

PheonixPharts··on Fix: Hong Kong is not in China
Prosperity is the most powerful propaganda of all.

I think few (non-Chinese) people in the US/Europe really understand the dramatic changes that have happened in the last few decades. I have Chinese family members that grew up under literally starvation conditions, only to grow up to find themselves multimillionaires. When Deng Xiaoping went to Shenzen it was a basically a small fishing village, and he said he would make it prosperous and change China's future. It happened. China went from developing nation to second largest economic power in the world in a life time.

For context I am not Chinese and not a fan of the CCP, but the zealous patriotism of the Chinese is not radically different than the zealous patriotism of American's in the immediate postwar period.

When discussing propaganda with Chinese citizens, it's vital to recognize this first. They aren't brainwashed masses, they are largely people who were told outrageous promises by their government, and in fact government policy really made those things happen.

PheonixPharts··on Qdrant, the Vector Search Database, raised $28M in a Series A round
Can you name me a single 10k-30k people company that has their own internally built relational database? Their own internally developed document database? I've never seen this in my career.

I don't even know any 10-30k people companies that build their own search, most I've known use elastic search or lucene.

It's had to imagine that a few of these vectordb companies don't establish themselves as the standard solution, being the equivalent to MongoDB in their space. The other competent players will very likely get acquired.

Certainly these vectordb companies are in a better long term standing that the bajillion companies rushing to build products that are just calling an API endpoint at the end of the day.

PheonixPharts··on Teach Yourself Programming in Ten Years (1998)
> from agile, which 20 years ago WAS good.

It's completely wild to me to have been in the field long enough to remember when agile was revolutionary, and then to see it slowly become twisted by decades of mismanagement into something that people hate today.

For most engineers today "agile" just means using jira, 30-60 minute long "stand-ups", and a perpetual panic about not getting enough done in 2 weeks.

PheonixPharts··on Teach Yourself Programming in Ten Years (1998)
I've worked with a fair number of less experienced, ChatGPT focused engineers and they aren't all that different from the Java engineers of yore that were completely helpless outside of an IDE and could only work extending existing code. These latter engineers where the exemplar "blub" programmers from PG's famous essay [0].

But clearly using an IDE does not make one a bad programmer any more than using ChatGPT will in the future. The bigger issue is the field is awash in people not really interested in programming. There's nothing wrong with this as everyone has to make a living, but this has a far bigger impact on the quality of engineers out there than the nature of the tools used.

Not long ago I was at a hip Bay Area startup and I don't think any of my coworkers, senior or otherwise, spent a second of their free time thinking about programming. For me I program for a living because it's great to get paid to do my hobby during the day. Getting started in the field during the shadow of the dotcom bust, the majority of the senior engineers that inspired me were likewise obsessed with programming and would be programmers even if it paid minimum wage. I don't think I would have necessarily become a programmer today since I wouldn't have been near enough the flame to ignite my own spark.

0. https://paulgraham.com/avg.html

PheonixPharts··on I'm sorry but I cannot fulfill this request it goes against OpenAI use policy
> they have far better coordination

I think LLMs are still underutilized, but to this point, it's been repeatedly shown that even the most state of the art LLMs are incapable of generalization, which is very necessary for coordinating large scale conspiracies against humanity.

PheonixPharts··on I'm sorry but I cannot fulfill this request it goes against OpenAI use policy
I don't find this any different than seeing an exposed jinja template: "{{product_name}} is perfect people who work in {{customer_industry}}" or the typical recruiter "Dear {{candidate}} I read your profile carefully and think you'd be perfect for {{job_title}} because of your experience at {{random_co_from_resume}}"

If anything, I think it's kind of cool that we're seeing LLMs actually used for something very practical, even if it is spammy (I mean I don't think template engines are evil just because they make spam easier).

PheonixPharts··on Bayesians moving from defense to offense
I'm curious how much professional statistical experience you have? In my experience probabilities alone are very rarely used and distributions of beliefs are combined with distributions of value so you can choose optimal risk/reward scenarios.

Bayesian decision making typically involves combining a cost function with your posterior distribution.

It's worth pointing out that this is typically not possible in most frequentist frameworks since they work in double negative style assertions and arbitrary threshold rather than modeling the problem itself.

While you can map Bayesian approaches to frequentists tests, this is usually a misunderstanding of Bayesian methods coming from a purely frequentist background. Bayesian analysis is fundamentally more flexible since it just the application of the sum and product rules (you don't even need Bayes' theorem since it can be trivially derived from these) as well as corresponding cost/value functions.

PheonixPharts··on Bayesians moving from defense to offense
I know this reply is a bit late, but, at least in my professional experience, Bayesian decision making typically does not involve just a probability of an event (in which thresholds are more or less arbitrary, including p-values) but involves mapping the probably to a cost/value in which the threshold isn't determined by arbitrary probability threshold but died to business value.

In the most common situations this typically means the expected value well exceeds the expected cost and you're comfortable with the risk if you're wrong.

PheonixPharts··on Bayesians moving from defense to offense
Bayesian statistics is fundamentally less complicated than Frequentist statistics since everything can be derived from a very simple set of first principles, rather than complex frameworks of ad hoc testing methodologies.
PheonixPharts··on Bayesians moving from defense to offense
> But I don't see p-values as connected to frequentist statistics.

P-values are fundamentally Frequentist because this framework see that observed data as random, whereas Bayesian statistics believe our observations are the only thing that is actually known, and all the other parameters are what is random in our experiments. That is: the data is the only part of an experiment that is real, everything else that we can't directly observe is something we must hypothesize about.

PheonixPharts··on Bayesians moving from defense to offense
In Bayesian terms there is no distinction between hypothesis testing and parameter estimation.

If your hypothesis is that A is greater than B, then you're test boils down to "The probability that A is greater than B", which is arrived at through parameter estimation via Bayes' Theorem.

You can consider the distribution of estimates for a parameter to represent a space of possible hypothesis for the true value of that parameter and their absolutely or relative likelihood based on the information available.

The also has the pleasant consequence that hypothesis testing using Bayesian methods nearly always is closer to the actual question someone wants to answer rather than replying with confusing statements such as "the probability of rejecting the null hypothesis", which contains an implicit double negative.

PheonixPharts··on Nvidia CEO: We bet the farm on AI and no one knew it
> just found themselves in a lucky situation.

Then this luck should have equally found AMD, who even today are struggling to pick up the ball they've been dropping for a decade now. My last PC had a Radeon, and I waited the life time of that PC assuming AMD support was just around the corner, all the while renting Nvidia cards in the cloud for any serious projects.

I've been in the ML space long enough to remember when people were just speculating about doing ML/computation on GPUs. Nvidia made that much easier and has continued to improve support and features for the past decade+ There insane success is certainly part luck, but I wouldn't be so quick to dismiss all of it as merely happenstance.

PheonixPharts··on Does quantum theory imply the universe is preordained?
Can you clarify what "true randomness" means out side of the context of human agency and perception?

Can you also give an example where "true randomness" would differ from the randomness I describe on the actions a human would take? That is to say, if a my internal model of the world tells me a coin landing on heads is 50/50 how does "true randomness" or not of the resulting coin toss impact how I should gamble on coin tosses?

imho the idea of "randomness" separate from our own internal state of belief is a nonsensical concept, but would appreciate examples to the contrary.

PheonixPharts··on Does quantum theory imply the universe is preordained?
Does believing that gravity exists mean a person will not roll down a hill?

Why would belief that we are just agents responding predictably to stimuli imply that one would not respond predictably to stimuli?

The fact that ones behavior does not change when one acknowledges that all behavior is deterministic and predetermined seems evidence in favor not against that hypothesis.

← PreviousPage 5 of 15Next →