HNHacker News
TopNewBestAskShowJobs

PheonixPharts

4,161 karma · joined April 7, 2022

submissionscomments
PheonixPharts··on Banana prices to go up as temperatures rise, says expert
Cocoa is having an even more dramatic rise in price[0], blasting through record highs right now, likewise related, primarily, to climate change.

0. https://tradingeconomics.com/commodity/cocoa

PheonixPharts··on Ask HN: Who's getting their job applications rejected?
What kind of organization culture do you have where someone leaving after two years causes an "amount of pain" and you have to spend considerable resources on-boarding new hires? That sounds like a deeper problem to me.

I tend to work pretty short stints at places, but in the vast majority of cases am shipping things to prod the first week or so and building and releasing revenue generating products in the first 6 months. Most of the places I've left, even after a short tenure, are continuing the work I started just fine and generally get lasting value from a even a year of my time.

With few exceptions, most of the teams I've left are sad to see me go, but that doesn't diminish the befit my time there has provided.

On the other hand most of the people I know that have worked 7+ years at a place aren't particularly engaged in the products and certainly aren't keeping up with major industry trends/changes/etc, and generally just go to work each morning without leaving much of an impact. Even at places like Google, most of the 10+ year people (with some notable exceptions) are the "rest and vest" types.

PheonixPharts··on Ask HN: Who's getting their job applications rejected?
> I’m in a mild panic about my future career despite knowing I’m capable. I may end up using my CPA to switch back into accounting for the time being

Having started my career in the shadow of the dotcom bust, I can say that this is exactly what most technical workers did. I started my career outside of tech and most of my grad school classmates were people trying to find a new career outside of software.

Many of the people I knew in this situation who were truly passionate about software ended up eventually returning, but quite a few others found the alternate career more satisfying in the long run.

The problem right now is that nobody else has really abandoned their tech career, still too early with too much money on the line. This means the signal to noise ratio for hiring is abysmal. If this keeps on long enough that software is no longer cool (and more importantly, not as high paying), it will be easier to reenter the industry.

Till then the only people that won't be struggling to find work are people with good networks and/or really standout backgrounds.

PheonixPharts··on Ask HN: Who's getting their job applications rejected?
I'm guessing your career started during the asset bubble (that we're arguably still in)?

Historically speaking, just because you "solved the problems" doesn't mean you "get the job".

When I first started working (not originally in tech, but plenty of tech places did this too) this is how (professional, skilled) jobs were filled:

- Position posted with a deadline for submissions (usually 2 weeks to a month out, sometimes more).

- Organization waits until all applications are received.

- Resumes/Cover letters are reviewed.

- Calls to screen candidates.

- 1/2 day to full day onsite scheduled (depending on level of role).

- 1 candidate gets selected from the pool who made it to the final round. It was also not uncommon that the organization would decide that none of the candidates had the qualifications they were looking for and they would simply, wait and start the process over in 3 months.

It was much closer to buying a house in a hot sellers market than tech hires in the last decade.

This world of continuous hiring where the goal is to maintain a headcount rather than fill a role is fairly exclusive to the recent tech boom where growth was what mattered most. It's historically normal to get to the final onsite and still lose the offer because competition beat you out.

PheonixPharts··on Daniel Kahneman has died
What's wild to me is that anyone could read chapter 4 and not look up the original papers in disbelief.

Long before the controversy was public I was reading that book and, despite claims that the reader must believe the findings, it sounded like nonsense to me. So I looked up the original paper to see what the experiment set up was, and it was unquestionably a ridiculous conclusion to draw from a deeply flawed experiment.

I still never understood how that chapter got through without anyone else having the same reaction I did.

PheonixPharts··on DBRX: A new open LLM
While GPUs are still the kings of speed, if you are worried about VRAM I do recommend a maxed out Mac Studio.

Llama.cpp + quantized models on Apple Silicon is an incredible experience, and having 192 GB of unified memory to work with means you can run models that just aren't feasible on a home GPU setup.

It really boils down to what type of local development you want to do. I'm mostly experimenting with things where the time to response isn't that big of a deal, and not fine-tuning the models locally (which I also believe GPUs are still superior for). But if your concern is "how big of a model can I run" vs "Can I have close to real time chat", the unified memory approach is superior.

PheonixPharts··on Fewer U.S. scientists are pursuing postdoc positions, new data show
I'm not sure why this is being down voted.

The classic argument in favor of taking a (potentially major) pay cut to work in academia as opposed to industry was always that academia is an environment for true learning and doing proper research uncorrupted by profit-seeking motives of corporations. Essentially, you make the classic trade-off between income and integrity.

In recent years we've seen that there's not a whole lot of integrity in academic research. Most of it is politics, grant hunting and working to inflate someone's ego; integrity of the actual science being done be damned.

At least with the current state of industrial research your results have to pan out enough to convince the market. Maybe LLMs are a overhyped, but they at least work enough to impress the people using them. Meta might never create the metaverse Zuckerberg dreams of, but at least people doing research on the hardware will hold in their hand lighter, higher performing, and truly state of the art VR headsets.

PheonixPharts··on Show HN: I made a cheap alternative to college-level math & physics tutoring
As someone who has used linear algebra, statistics and calculus as part of my day-to-day work for years, I would be very cautious of relying on ChatGPT as your "tutor".

Occasionally I've tried to substitute ChatGPT for the shelf of reference books in my office and nearly always had poor results.

The trouble is that the outputs of these models, by their very nature, look convincing. Even as an expert it takes a fair bit of background to realize when ChatGPT is making a mistake. The results virtually always look good at first pass.

I would strongly recommend sticking to text books for self study.

PheonixPharts··on How Chain-of-Thought Reasoning Helps Neural Networks Compute
Given that LLMs are basically doing Sequential Monte-carlo sampling in latent space, the "thought" part of chain-of-thought certainly seems more akin to the necessary warm up period whenever you do any kind of SMC sampling.

Anyone whose done serious Bayesian stats work knows that the sampler needs to warm up for a bit to get start efficiently sampling. I suspect something similar is happening with chain-of-thought: the model needs to wander around a bit before it gets into the correct neighborhood for sampling the answer.

PheonixPharts··on The Reddits
Reddit IPO'd today and the purpose of this piece is to add whatever hype can be added to that IPO so that PG and other early investors can maximize their return, and then likely, slowly and carefully unload their positions into the public and other investors who will be the bag holders when the price inevitably declines over time.

The "idea" is "make PG even more money", and that's a great idea if you're PG.

PheonixPharts··on Is the Job Market Dying?
This is not true. The one benefit of all the AI hype is that there are many small teams of smart people getting funding. Maybe in 2-5 years they'll all get acquired or go under, but right now is some of the most fun I've had in industry in a long time.

And, despite all the hype, there is a lot of unexplored stuff in this space that can be done on hardware that can be run in a home office.

Yes, it's not literally coding apps any more, but I know quite a few people working on small, funded teams having a lot of fun right now.

PheonixPharts··on Is the Job Market Dying?
> The rest left tech for good.

This is an important reality check for many people.

I entered tech in 2004, during the immediate aftermath of the dotcom crash. At that time software engineers got paid well, but not insane. Additionally, every software engineer I met was passionate about writing code.

Honestly, around 2019 I was really missing those days. The field has become flooded with people looking for a high paying job with essentially no interest in software or computer science beyond the paycheck it provides. I don't blame people for wanting to make money, but I do miss virtually everyone in the industry being genuinely fascinated with software and programming.

The good news is, if you're the kind of person who would write code even if it paid minimum wage, you'll survive this. People whose book shelves are filled with CS books, who find themselves working on coding problems at night because it's fun, who can't help hacking around with new ideas on the weeked, will very likely continue to work in software. You'll likely make less money, but you'll also have more fun.

Unsurprisingly, in this current market I'm getting paid less but having more fun at work than I've had in nearly a decade.

PheonixPharts··on What would an LLM OS look like? (2023)
> The future will be agentic.

Minor pet-peeve, but I can't take anyone who uses the term "agentic" seriously. Computer Science has long used the concept of an "agent" and the adjective form of such systems has been "agent based" [0].

There is nothing that makes "agent based" LLMs different than other "agent" systems in computer science, so using the term "agentic", for me at least, is just short hand for "I'm not familiar with the history of AI in computer science".

I don't know if Karpathy has used that exact phrase, and if he has, he obviously is an expert in both AI in particular and computer science in general. But the majority of people I've heard use the phrase "agentic" are just trying to sound cool/cutting edge when referring to a concept that has a long history in the field.

0. https://en.wikipedia.org/wiki/Agent-based_model

PheonixPharts··on Mass timber is great, but it will not solve the housing shortage
> NIMBYs worry that the mere existence of nearby multifamily will decrease their home value. In rare and unsympathetic cases ... that may be true.

It's not rare. I've known multiple people, myself included, who have decided not to make an offer on a single family home because it neighbors a gross apartment building.

The vast majority of home buyers want single family homes. I don't know where HN obsession with vertical housing comes from, maybe because the majority are young and still live in apartments, but this isn't reflective of most home buyers

The only adults I know who will even consider purchasing a condo are single with no pets and no intention of having a family. Even among most childless couples, once they live in a single family home they would never go back.

PheonixPharts··on Devin: AI Software Engineer
It's worth pointing out that on their eval set for "issues resolved" they are getting 13.86%. While visually this looks impressive compared to the others, anything that only really works 13.86% of the time, when the verification of the work takes nearly as much time as the work would have anyway, isn't useful.

The problem with this entire space is that we have VC hype for work that should ultimately still be being done in research labs.

Nearly all LLM results are completely mind blowing from a research perspective but still a long way from production ready for all but a small subset of problems.

The frustrating thing, as someone working in this space awhile, is that VCs want to see game changing products ship overnight. Teams working on the product facing end of these things are all being pushed insanely hard to ship. Most of those teams are solving problems never solved before, but given deadlines as though they are shipping CRUD web apps. The kicker is that despite many teams doing all of this, because the technology still isn't there, they still disappoint "leadership". I've personally seen teams working nights and weekends, implementing solutions to never before seen problems in a few weeks, and still getting a thumbs down when they cross the finish line.

To really solve novel problems with LLMs will take a large amount of research, experimentation and prototyping of ideas, but people funding this hype have no patience for that. I fear we'll get hit by a major AI winter when investors get bored, but we'll end up leaving a lot of value on the table simply because there wasn't enough focus and patience on making these incredible tools work.

PheonixPharts··on Among the A.I. doomsayers
I use these models every day, and have for quite awhile. More specifically the vast majority of my work on them has been evaluating their performance.

While there is certainly untapped potential with this models, what we've seeing is unquestionably market driven hype.

I know very, very few people who are working closely with these models who buy the hype.

PheonixPharts··on Show HN: Prompts as WASM Programs
> That's problematic because the entire point of an LLM

Could you detail what the results of that problematic behavior are? Do you suspect structured generation performs worse than unstructured?

> squishing the result into a json schema

The strategies I've seen for dealing with structured generation seem to be ensuring that the structured is represented in the prompt, so the model isn't squishing anything.

> For longer text fields, many SOTA LLMs are prone to close the field early and append an ellipses rather than flesh it out in its entirety. Constrained schema generation will have a high probability of closing the (broken) string and a low probability of continuing past the (incorrect) ellipses,

Part of structured generation is specifying exactly the range you want for the field, so I'm not sure how this issue arises. With structured generation you force the LLM to have at least some number of characters, so this seems like an argument in favor of structured generation rather than against. If unstructured LLMs want to "close the field", then structured can force it to remain open.

PheonixPharts··on Show HN: Prompts as WASM Programs
> The most obvious usage of this is forcing a model to output valid JSON

Isn't this something that Outlines [0], Guidance [1] and others [2] already solved much more elegantly?

0. https://github.com/outlines-dev/outlines

1. https://github.com/guidance-ai/guidance

2. https://github.com/sgl-project/sglang

PheonixPharts··on Google fires employee who protested Israel tech event, shuts forum
Are you being deliberately obtuse? Surely you recognized there is a context to the "no interrupting meetings" guideline. The protester felt that meeting was playing part in the support and facilitation genocide, I think that's adequate grounds to interrupt a meeting.

And what do you find absurd about the "slave owner" comparison? I'm not arguing that modern day employment is slavery or anything of the sort, just pointing out that the terms of "reasonable expectations" are always defined by those in power, and not once in history has working within "reasonable expectations" caused a shift in those power dynamics.

PheonixPharts··on Google fires employee who protested Israel tech event, shuts forum
If you earnestly believe there's no important distinction between employees of a multinational corporation and a guest in your private home then I don't think there's any more possible room for discussion.

I am curious though, in what specific context would you "defend to the death" my right to free speech? It seems strange to be willing to offer your life for something with so many terms and conditions applied.

PheonixPharts··on Google fires employee who protested Israel tech event, shuts forum
This same logic can be used to justify any system of exploitive power. Companies control my ability to eat and be housed, these companies drive global politics. If the voice and ethics of the people are not allowed to influence these, then what voice to do the people have? Is my choice is to keep in line or starve how is that any different than any other despotic system?

You're own profile says "I may disapprove of what you say, but I will defend to the death your right to say it." but clearly this only applies for the 8 hours of the day I am not at work (less if I work late)?

PheonixPharts··on Google fires employee who protested Israel tech event, shuts forum
The trouble is that "reasonable expectations" are set by those in power. Slave overs set "reasonable expectations" for slaves, feudal lords set "reasonable expectations" for peasants.

In such a world "reasonable expectations" by their nature foster a continuation of the existing power structures. Behavior that challenges "reasonable expectations" does so precisely because of the belief that the continuation of the status quo as perpetuated by those in power has become untenable.

Many people "reasonably expect" nations and companies to not be complicit in genocide, where is their recourse?

PheonixPharts··on Google fires employee who protested Israel tech event, shuts forum
But how he did it is the important part.

I mean Aaron Bushnell could have written angry letters to his representatives, but he didn't. He lit himself on fire outside of the Israel embassy after declaring he would not take part in genocide and screamed "free Palestine" with his dying breath. And because of that, today I am writing about what he had to say.

If you think the correct response to genocide is politically discussing it with management then I don't think you really understand the gravity of the situation.

PheonixPharts··on Google fires employee who protested Israel tech event, shuts forum
What that means is "only status quo opinion and politics". It's implicit support of the mainstream ethos of corporate America, one which has well established it values profits over people.

Management works to serve those corporate interests, so it is in their interest to convey the message that the appropriate thing to do is be silent and comply.

PheonixPharts··on Google fires employee who protested Israel tech event, shuts forum
> propaganda news outlets.

Most major news outlets are still wildly pro-Israel. My and most people's opinions on the subject tend to not to drive from main stream media accounts, but research and journalistic investigations into the situation there.

I don't personally know anyone who has walked down the path of doing real research into the issue and not come out of it horrified with Isreal's actions and this has been long before Oct, 7 2023. Most people who are "fine" with the situation are the ones getting their opinions from the news.

PheonixPharts··on Google fires employee who protested Israel tech event, shuts forum
I'm confused by your logic here, the fact that all other big tech companies would do something is hardly grounds for considering an action "right" in any moral framework I know of.

"Unsurprisingly so" I can get behind. But "rightfully" is a stretch I wouldn't agree with.

PheonixPharts··on Four new models that are benchmarking near or above GPT-4
The benchmarks are so easy to game it's not meaningful unless you see a major difference and even then it can be questionable. For example, did anyone read the Gemini paper[0]?

The team details the slight modification they made to the metric which is basically a tweak on maj@32 that worked in their favor. They choose to do the majority vote of 32 samples unless the probability of the highest sample was too low and then they chose greedy, they also had to learn what "too low" was. Sound convoluted and a bit contrived? They even show plots that clearly demonstrate they settled on the metric because it's the one where GPT-4 results remain they same, but Gemini wins.

Don't get me wrong, LLM evaluation is essential. But in order to do this you need to be very consistent between comparisons and even then the results can be misleading.

0. https://arxiv.org/pdf/2312.11805.pdf

PheonixPharts··on Gabriel García Márquez: Sons publish novel that late author wanted destroyed
> Do we get to look at your personal journals, browser history and hidden photo gallery once you kick the bucket?

Once you are dead, you are dead, it's the same as if you never existed in the first place. I completely fail to understand how my wishes are relevant after I no longer exist.

What shame or embarrassment can I feel when I'm dead? Promise me on my death bed my secrets will never be revealed and the second I'm gone, share them and it won't change the peace of mind I had. Likewise the state of my mind at the moment of death is irrelevant because I'll be dead.

It's bizarre to me to see you are so troubled by this. Do you care what happens after your dead? Do you care what's happening on a planet so distant and foreign you do not even know it exists or what it's inhabitants are like? The world after you are dead will be far more foreign to you than such a world that exists today.

In time all humans will be dead. In time all evidence of this planet will be erased. If alien life should come across some rock with Gabriel García Márquez last writing and note saying "do not read" are found should that alien life honor that request? What moral system are you deriving these beliefs from?

PheonixPharts··on Worldcoin hit with temporary ban in Spain over privacy concerns
It's sort of like the twist at the end of Soylent Green. The real twist isn't that "Soylent Green is people!" the twist is that you can see this with your own eyes, shout it in the street, but ultimately no one really cares.
PheonixPharts··on Worldcoin hit with temporary ban in Spain over privacy concerns
I was chatting with a few Chinese crypto people the other day, and the general consensus in that community seems to be that Worldcoin basically operates as an ATM for Sam Altman.

He undoubtedly owns a considerable amount, and it doesn't matter if it's a long term successful so long in the near term it has a multiplying impact on his OpenAI work. If you look at the 1 year change in WLD you see that he's likely made a pretty sum from it already, especially since whatever amount of WLD he own was acquired at effectively zero cost[0].

The people I was chatting with shared a similar view to dogecoin and Musk. All either needs to do is drive up a little hype and in to time they have access to tremendous amounts of liquidity.

0. https://coinmarketcap.com/currencies/worldcoin-org/

← PreviousPage 3 of 15Next →