HNHacker News
TopNewBestAskShowJobs

mediaman

6,587 karma · joined April 20, 2009

submissionscomments
mediaman··on First Impressions with GPT-4V(ision)
You'll have it within a week or so.

Pretty much all new products that require significant per-user incremental workloads (e.g., in this case, significant GPU consumption per incremental user) do rollouts. It's an engineering necessity. If they could roll it out to everyone at once, they would.

mediaman··on So You've Decided to Move from Unity to Unreal Engine
I get the impression that Unity did not do a lot of homework before rolling this out. Several aspects of the policy they’ve waffled on, and when asked about who would pay the fee for GamePass games, they said the distributor would, but it appears the distributors were not consulted on this.

Not to say they didn’t consult with the major Unity players, but it seems plausible that they didn’t based on the carelessness of the rest of the rollout.

mediaman··on ChatGPT Enterprise
Pretty unlikely. Generally you don't use fine-tuning for bare documents. You use retrieval augmented generation, which usually involves vector similarity search.

Fine-tuning isn't great at learning knowledge. It's good at adopting tone or format. For example, a chirpy helper bot, or a bot that outputs specifically formatted JSON.

I also doubt they're going to have a great system for fine-tuning. Successful fine-tuning requires some thought into what the data looks like (bare docs won't work), at which point you have technical people working on the project anyway.

Their future connection system will probably be in the format of API prompts to request data from an enterprise system using their existing function fine-tuning feature. They tried this already with plugins, and they didn't work very well. Maybe they'll come up with a better system. Generally this works better if you write your own simple API for it to interface with which does a lot of the heavy lifting to interface with the actual enterprise systems, so the AI doesn't output garbled API requests so much.

mediaman··on A venture-backed startup has quietly bought more than 80 mom-and-pop shops
Usually it’s more structural than that. Too much competition, too few customers, too little scale to afford the type of talent that can optimize inefficiencies. There is a reason these small localized businesses tend not to produce outsized profits.
mediaman··on Code Llama, a state-of-the-art large language model for coding
That model is trained on synthetically AI-generated code, not internal code.

It suggests that synthetic training could be the future in increasing capability of smaller models (and perhaps bigger ones too). AI will train AI.

mediaman··on Don’t build a general purpose API to power your own front end (2021)
If you use an API, but in a way the author suggests (conformed to the UI, rather than an abstraction), what's the simpler/better way to do auth than API tokens?

I have depth of experience in a niche industry and I'm working to build solutions for that industry, but I'm still early on the learning curve of being an effective full stack developer to make those solutions.

mediaman··on Cloud Provider Gets $2.3B Loan Using Nvidia's H100 as Collateral
Lease deals like this tend to be structured as an amortizing loan over 3-5 years. They also probably did not finance the entire value, but the entire value will be collateral. If they financed, say, 70%, then a couple years in the remaining exposure may be well under 50% of original price; four years in, they likely need only a low recovery to become whole.
mediaman··on Cloud Provider Gets $2.3B Loan Using Nvidia's H100 as Collateral
I don't know the details of this lease, but it is likely a capital lease, not an operating lease. The lessee will not have the option of returning it. It will be recognized as an asset on the lessee's balance sheet. The lessee backs the lease not just with the collateral, but with the full value of the capital structure up to any debt senior to it.

It's a bet against progress in the field only to the extent that they depend on the collateral if the lessee fails. Lessors hate getting collateral back. So they've probably looked at the lessee's capital structure and decided that, along with whatever interest rate they're getting, it's a reasonably priced risk. They're likely mapped out their estimated risk of default by year along with a declining recovery value of the collateral by year.

mediaman··on Run every day (2021)
Right. As if half marathons are supposed to be something in a wholly different universe than running a mile a day? It's great exercise, and I personally wouldn't want to do more than a half marathon, since the health benefits seem to be kind of limited and injury rates start creeping up. But someone who runs a mile a day is in reasonable shape, and the distance in fitness from a mile a day to running a half marathon isn't so great.

In contrast, I know people who do ultramarathons, running 50+ miles at a stretch, and I simply cannot imagine it. I'm impressed with people who do: the training, the determination. You can't just wing that. You can't do it on a whim. A half marathon you can do a half-ass job preparing for, and it won't be fun, and you'll feel bad afterward, but it'll be doable. Different categories.

mediaman··on US inflation cooled in June for the 12th straight month
I think it's unfair to pin this on the Fed.

Most of the inflation appeared to come from a mix of massive fiscal stimulus and supply chains that got blown up by the pandemic, so large increases in demand coupled with decreased capacity to supply, driving up pricing for capacity.

Arguably increases in some asset prices could be attributed to QE, but the story doesn't really hold water for CPI ex-housing.

As an aside, I don't even think this course of events was a mistake. It was much better to have some inflation from huge stimulus but avert economic disaster, and then deal with the inflation through the relief valve of higher interest rates. There is no relief valve for total demand destruction.

mediaman··on StabilityAI cofounder says CEO tricked him into selling stake for $100
Stories like this are bizarre. There are a million ways to ensure rights for a founder-owner that do not depend on 51% ownership, such as requiring supermajority votes for replacing the CEO, or right of first refusals granted to the founder-owner for share transfers. If they throw a fit about those terms, then don't do the deal!

If you are selling shares to a PE firm with the explicit goal of retaining control, and you have less than 50% of shares, why would you make it so easy for them to get control?

mediaman··on GPT-4’s secret
The author chooses whether to make it free or not. It’s not like medium.
mediaman··on Workers with less experience gain the most from generative AI
Every time someone posts something like this, it’s amusing because they clearly don’t try it in GPT4 before they post.

GPT4 nails your question about going east from New York in a boat, even though I think that’s a very poor test of ‘intelligence.’

I have no idea what you’re asking with your second question (what city are you talking about? There is no context), so I guess I, too, am not intelligent.

mediaman··on Bard is getting better at logic and reasoning
I wonder why they don’t just charge for it.

Release a GPT-4 beating model; charge $30/mo.

That’s not aligned with their core ad model. But it’s a massive win in demonstrating to the world that they can do it, and it limits the number of people who will actually use it, so the hardware demand becomes less of an issue.

Instead they keep issuing free, barely functional models that every day reinforce a perception that they are a third rate player.

Perhaps they don’t know how to operate a ‘halo’ product.

mediaman··on Notes on Vision Pro
They have a training process at the beginning that learns your eyes. It asks you to ‘point’ to various points in space. I’d imagine it would be able to detect a dominant eye with that, and prefer it over a slower or non-tracking eye.
mediaman··on What Neeva's quiet exit tells us about the future of AI startups
I believe there is better opportunity from focusing hard on specific niche industry verticals and developing AI assisted workflows that make their operations more efficient.

Those will be good businesses, and founders will do well, though they may not be billion dollar businesses. Their moat will be industry comprehension and then their integration and embedding into business operations.

That does leave a question as to the purpose of VC in AI though.

mediaman··on AI Canon
a16z's investment thesis completely centers around finding the next bagholder for their investments.

They don't care about building enduring businesses that make them a lot of money because the businesses are actually driving that much value creation.

They care about creating a hype cycle and dumping before the tide goes out and their investments crash on someone else's balance sheet.

None of them could get on a podcast and say anything remotely rational or intelligent about their crypto investments: it was all pump and dump. In one interview, their head of crypto investments main value claim was "monetizing the sharing of your home wifi," and after some pushback resorted to deference to authority by saying the founders they invest in went to the right schools and nobody else could understand because "they weren't in the room."

These are smart people who couldn't articulate a clear value. They obviously didn't believe any of it and just saw an opportunity to promote a Dutch tulip mania.

Good fund for con artist "entrepreneurs" though. Same deal for Chamath and his SPACs.

There's increasing hesitation among builders to take money from funds that have strayed this far from long-term value creation, because once they get addicted to short-term pump and dump profits and chasing the latest thing, it's hard to go back to supporting actual builders.

mediaman··on Intel Announces Aurora GenAI, ChatGPT Competitor with 1T Parameters
It's a typo in the article. GPT-3.5 is 175bn parameters, not 175 million.

GPT-4 does not have publicly available parameter count, but is thought to have something similar to 1tn.

mediaman··on Americans have never been so unwilling to relocate for a new job
You are generally right, but currently most homeowners have mortgage debt at rates far below market, so they could afford much less house than before (if they need financing).
mediaman··on Don't believe the hype: why ChatGPT is not the “holy grail” of AI research
Not understanding why this is an issue for LLMs but not humans.

This is a simple commercial decision to make governed by three factors.

1. What is the cost of making an error?

2. What is the cost of the human doing the work?

3. What is the likelihood of the human making an error?

It's just evaluating how much more likely AI is to make an error than a human, by the cost of that error, set against the savings by using fewer humans.

Look at the legal profession. Sometimes the cost of an error is high, but usually it is not. There are already tons of little errors in contracts and discovery, and today they're all human. And people are very expensive. There is a giant swath of legal work that looks very attractive to automate at less than 100% accuracy.

Customer service: people offer poor customer service all the time, and usually the cost of that error is low. Human customer service isn't as expensive as legal work, but it's still relatively expensive. Very attractive to automate at less than 100% accuracy.

mediaman··on Don't believe the hype: why ChatGPT is not the “holy grail” of AI research
The mistake is in believing that LLM's output should be deterministic to be useful.

Human output is not deterministic.

Fields with text-heavy output are already being upended by this. Being able to summarize long legal briefs, identify contract problems, do classification of discovery documents, or even write first drafts of common legal forms is already upending the legal discipline.

Chat-based customer support agents are seeing 25% productivity improvements based on two-year-old models for new employees, according to a study published in NBER.

Things like BabyAGI and other sequential "do anything" tools appear to be close to useless now, and unfortunately that is what is catching a lot of hype on Twitter. But actual industry applications are much quieter (often NDA) and much more impactful.

mediaman··on Don't believe the hype: why ChatGPT is not the “holy grail” of AI research
I've built internal systems that do summarization based on knowledge retrieval systems for specific nonpublic corporate information.

With GPT-4, I find very little hallucinating. It very rarely deviates from the source material. Every time I've found something unexpected, there was a problem in the source material provided to the model.

mediaman··on Don't believe the hype: why ChatGPT is not the “holy grail” of AI research
More on the port workers: today you'll find that skilled union port workers are extremely well-paid. Workers in the cranes that pull containers off the ship, for example, are paid >$300k.

They are quite skilled, but so are many other skilled tradespeople. Electricians and machinists for example do not make that kind of money despite their skill. So why do crane operators make so much?

When containerization, and subsequently other types of automation, hit ports, the union resisted fiercely but ultimately had to work out a deal with the port operators.

That deal was to reward the most tenured union tradespeople with much larger pay packages, at the cost of the less experienced tradespeople, who would have to find different work.

This agreement became agreeable to both sides, since the operators could still massively reduce the workforce, and the only price they'd have to pay would be high salaries for the union workers who remained. And the union was able to reward their longest tenured members.

It's hard to fault the union for this: the alternative was likely both huge reduction in workforce and less attractive pay, so they at least got good pay out of it for the remaining workers. But it made a lot of the less tenured union workers resentful because they felt the union sacrificed them in favor of the union 'insiders'.

The book 'The Box' by Marc Levinson is a great coverage of this topic, if steel shipping containers are the sort of thing that get you going.

mediaman··on The rise of ChatGPT-enabled GitHub spam
The answer, of course, will be to include insults in any message as a signature of authentic humanity.

This will work until non-RLHF models become good.

mediaman··on It looks like GPT-4-32k is rolling out
It does not have memory outside the context window. If you want to have a back-and-forth with it about a document, that document must be provided in the context (along with your other relevant chat history) with every request.

This is why it's so easy to burn up lots of tokens very fast.

mediaman··on Samsung bans use of A.I. like ChatGPT for employees
ChatGPT does use input as training.

The OpenAI API has a policy to not use input as training data.

Azure also offers the API with an agreement not to use it for training.

mediaman··on ChatPDF – Chat with Any PDF
It is returning a series of snippets from the book, from which it cannot create a summary. That’s because of the way vectorized search works.

Summarizing the book requires a different approach. Usually condensing the book, maybe processing ten pages at a time, and then summarizing the condensed chunks.

mediaman··on Americans’ IQ scores are lower in some areas, higher in one
Sounds like it was a high pass filter, and that IQ was a necessary but insufficient condition. Or they just had too many applicants.

For what it's worth, g-factor is correlated to many different cognitive abilities. A person's individual competency areas are highly correlated to other competency areas. For a job where cognitive ability is relevant, knowing the g-factor is predictive of job performance.

mediaman··on Everybody is the main character
This is laudable, and you sound like someone who is both competent and works with competent people. In those cases they should not be micromanaged because it will stunt their growth and result in worse outcomes because of the greater knowledge the IC will have for the task at hand than the person not doing the task.

But those preconditions of competence are not always in plentiful supply.

Some team members must be empowered, be given a target to go after, be given a domain of responsibility, and then let loose, with little interference except to assist in politicking or obtaining organizational resources.

There are others, however, where this will be disastrous. Frequent errors, stopping progress without raising a flag, inability to adjust course and learn, inability to be resourceful, lack of agency, inability to map objectives to tasks required. Experienced managers have all had this kind of team member, and recognize after many attempts that "be a better leader" often doesn't fix it.

The easy answer is to say, "well, simply don't hire this latter group of people, or PIP them if you do." But they exist, and they will wind up working somewhere, even if perhaps they are not working at a large, extremely well-paying tech company that typically will have plenty of applicants for their RSU-laden opportunities.

So, yes, those classic leadership qualities are great when the team member will thrive with them. In those cases the leader must provide the target, the support, and 'get out of the way.' But it is the rather ugly reality that this is not always the case. And in those, the leader must adapt their approach.

mediaman··on Everybody is the main character
It's inspired from literature. It means someone who views themselves as the main character of the world's story. In some literature, the world tends to revolve around the main character. (This is not true of much of literature that deals with, for example, character studies, but think of books like Ender's Game or Lord of the Rings.)

Although we are each usually the 'main character' of our own story, we typically know that we are not "the main character" of the world broadly. Some people act in ways that suggest they may not understand the difference. People refer to them as having "main character syndrome."

← PreviousPage 11 of 34Next →