andrewcamel
andrewcamel on Hacker News
Re per token, I had the same reaction, but given both labs are economically advantaged moving customers to per-token consumption... almost want to avoid this on principle. Even if not intentional, benefitting from a degraded product is not something I want to accept or enable.
More now than ever (since original ChatGPT release), the OSS models and open harnesses (eg Pi) are looking mighty attractive.
I’m not sure it will cause a reversal, but I bet the k8s takeover wouldn’t have happened with costs of today. You inherently end up using far more resources than you need - between HA, built in services, overhead per node, etc.
I personally will be using more resource efficient approaches in everything I do. Question is just what provides the closest set of benefits without the full k8s weight.
I’d imagine you could seed it more easily by focusing on top 50 routes by passenger count in domestic USA. Then go from flight schedules for top airlines into tail numbers into body types etc.
Big fan. One feature idea/request - a map showing coverage with 0-100% by route (red/yellow/green lines). I’m just curious to see where I should think to look for / expect starlink options. Probing into a few upcoming trips showed basically no coverage.
Philosophical question, but after reaching critical mass, should languages even aspire to more? I.e. do you risk becoming "master of none"? What's wrong with specialist languages? I.e. best of breed vs best of suite?
I agree with author Go is getting squeezed, but it has its use cases. "COBOL of could native" implies it's not selected for new things, but I reach for it frequently (Go > Java for "enterprise software" backends, Go > others for CLI tools, obviously cloud native / terraform / CI ecosystem, etc.).
However in "best of suite" world, ecosystem interop matters. C <> Go is a pain point. As is WASM <> Go. Both make me reach for Rust.
What response would you suggest for Israel?
My understanding is their civilian vs combatant ratio is lower than comparable urban conflicts. So it’s hard to respond “softer”.
Population adjusted, 10/7 was 10x+ worse than 9/11. Challenging to expect zero response.
So what else is there?
Are tariffs already in place or is this just a thinly-veiled scapegoat for haircutting traffic allocation by 95%? To a customer, it certainly feels like a bait and switch to sell a subscription product and once customers are embedded materially change the economic trade.
It's translation (text -> text), not speech -> text.
Do they report the % of revenue from AI / data center chips vs all others? Or is there a way to derive it? Surprising to hear the slump given what we're seeing in AI-driven GPU demand. I guess that's just not as meaningful a portion of the TSMC business as I would've thought.
It's a collection of interviews posted in the form of a library. So bit blog-like in structure, but just a collection of ideas on how to leverage this new tech.
I'm starting to outline them here: ctlresearch.com . Upcoming interviews with Chief Architect at Intuit, Head of Procurement at DoD, etc. DoD already shortened process of writing structured "requests from industry" from 3 months to 1 day. Makes it far easier to get requests out to vendors. Next step is an auto-complete bot that helps vendors respond with required language to RFPs.
I have 20 interviews coming down the pipe -- all of which have highly tactical / near term valuable ideas like this.
This is reminiscent of the USA vs Microsoft case in 2001... I wonder how much the antitrust team at Microsoft gets a say in product decisions like this. Just feels like they're toeing a line...
https://en.wikipedia.org/wiki/United_States_v._Microsoft_Cor....
Also definitely not in the best interest of users, which isn't the Satya Nadella way of operating, at least not as demonstrated in the developer tools side of the business.
Maybe not the right term... Just that a lot of other libs act like guardrails, i.e. let the model generate what it does (in full form text / GPT output), and try to then parse out what you want, error if output doesn't conform to standard format. As opposed to basically only allowing the model to generate into the already-built JSON form fields. Understandable why this guardrails/parsing approach is so popular though... can't do what this library is doing with OpenAI API. Need to be able to manipulate the token generation; otherwise you're forced to take full text output and try to parse it.
Seen a lot of things trying to do this by pressure testing the outputs, but all feel like anti-patterns. This is the first that seems like the "right" way to do it. Better to manage how the model is generating vs creating one more potentially faulty "glue" layer.
Very cool! Could definitely see myself using this. Out of curiosity, what's the visualization library you're using to generate the boxes/lines/zoom in-out canvas? Or is that all native to your app? Building something unrelated and looking for something like this
Vintage 2020-2024+ are going to be some of the worst for PE, after a 40 year bull run driven by rates going from 20% in 1981 to 0% in 2021. Now there's too much capital in the hands of people who have been trained for years that to be successful, all they need to do is deploy capital, without a real understanding of the underlying business. Just ask experienced operators if their PE board members know what they're doing; I'd venture to guess the vast majority would respond negatively. It will take a few years for this excess cash to get washed out, and then you'll have a longer-term washout of all these people who have spent years refining their skills to be good at capital deployment, not to return capital in a differentiated way (per Bain stat on 5% driven by profit). For so many years, the business has basically been "show up, be a reliable soldier / good politician, know how to deploy capital, and you'll get promoted and make a bunch of money". It's hard to see how that will remain the case, particularly as the LP return expectations go up transitioning from a positive sum environment to a zero sum environment.
It feels like we're in the middle of a "back to basics" moment. Funds will need to figure out ways to develop true operating advantages, but not in just a performative way like most have done in the last 20 years. Now it will really count. There are a small set of funds who are true experts, spend management fees on teams of experts that are constantly involved to support management (not just an outsourced recruiting team, a bench of "advisors" with nice brand names, or a Bain team on retainer, but truly know how to operate all aspects of a business). And the exciting thing is that there are people who do this well, consistently. I.e., it is possible with the right team and execution. And there is a particularly interesting opportunity right now in that so many businesses are PE-owned and poorly managed, for those who know what they're doing. Lot of things that worked before in a low/zero cost of capital environment will now break and create forced sellers.
Reality here likely ends up being far more complex / unpredictable, but directionally this feels accurate.
Incredibly well positioned vs WhatsApp, iMessage, etc in the new world of LLM-enabled agents / chat bots… I’d put money on others launching similar things. Hard to see the current web UI remain the long-term format on mobile, maybe web too if the Telegram / other messaging apps well support this.
On this topic, Apple is the sleeping giant. Sleeping tortoise maybe. Everyone else has been fast out of the gates, but Apple has effectively already been positioning to leap frog everyone after a decade+ of M1 chip design. Ever since these chips launched, the M1 chips have felt materially underutilized, particularly their GPU compute. Have to believe something big is going on behind the scenes here.
That said, wouldn't be surprised if the truth was somewhere in between cloud-deployed and locally deployed, particularly on the way up to the asymptotic tail of the model performance curve.
Big fan of this... suspect a lot of the modern SaaS BI / analytics players would benefit. Just end up hitting limits trying to use Javascript to render large datasets, having tried to build some of this myself. Happy to share some potential use cases / thoughts on where you could apply this if helpful! Feel free to reach out.
Would be incredible if the lab leak in China was intentional and they ended up saving the US by increasing the birth rate!
Yep... I bet this is about to pull forward a lot of startup death. Gun now to VC heads to save things. Forcing decisions they were hoping they wouldn't have to make for at least another 6-12 months.
Posted by
andrewcamel on Thursday March 02, 2023 @05:17PM
from the self-post dept.
Just a reminder to not store all your expensive eggs (embeddings) in one basket! Bad mistake on my part.Something to be said for Pinecone on this... if your core business is to be the "production-grade hosted vector DB", probably best to reallocate some of your R&D budget to implementing basic controls that prevent someone running a script that can wipe the entire production DB. Undermines...
Helps when you can limit OpenAI’s inference capacity during negotiations
What’s the lag time between launch orders and actual launches? Wonder if we’re looking at a 9-12 month trailing indicator of health and the broader tech reset is dampening orders in the last 2-3 months.
Speaking as someone who came from PE, agree that higher rate environment means both better investment alternatives and also more expensive capital for LBOs. That said, the firms with fresh capital can make a ton of money buying into this environment. Zendesk being a great first example. Not a huge amount of debt required to do these types of deals. Liquidity going to hide under the figurative rock of fixed income means values get dragged down, which creates an incredible buying opportunity for all these PE funds. Look 5-7 years out, I'd guess 2022 and 2023 vintage funds will be some of the best in history.
It’s challenging to actually buy a Raspberry Pi these days unfortunately. All the official resellers are out of stock of all the other models and resellers have these on eBay for 3-4x the list price. Maybe this one will be in stock for longer…
This is the type of investing one (at least me) dreams of doing if you're fortunate enough to get to $B+ in net worth. Sam knows enough people who knows how this tech works to get a read on quality / achievability, and if he gets it right (however challenging/unlikely), it means incredible things. If it doesn't work, whatever -- at least he's shooting at something interesting. And talk about something intellectually interesting to be involved with -- it must be a collection of great minds at this company.
I think it's easy to assume government agencies have weak / outdated tech, but the SEC is notoriously pretty advanced in how they identify bad actors. Think about situations where someone is passing information to a friend who does not work at a company and gets a cash kickback. Somewhere they're getting a graph of relationships, and then merging that with transactions + likely some detail around known public disclosures, and doing this all at a scale of hundreds of millions of people and billions[?] in daily trades.
Worth checking out Snowplow Analytics as a self-hosted version of Mixpanel. Good guys out of London that have largely bootstrapped it, but the business is pretty big now -- some good customers using it at large scale.