DebtDeflation

Probably worth thinking more about what we mean by "wrapper". A year or so ago, it often meant a prompt builder UI. There's no moat for that. But if in 2025 a "wrapper" means a proprietary data source with a pipeline to deliver it along with some proprietary orchestration along with the UI (and the LLM API being called), then it likely warrants looking at it differently.

The reason there was no political will to punish Airbnb and Uber for violating the law was that initially they were subsidized with VC money and so were able to undercut traditional hotels and taxis on price. In the world of tradable goods, pricing below cost with the intent of putting competition out of business so you can raise prices later is known as "dumping" and is itself illegal.

I'm a consultant and my team has been building interactive "sales agents" using tech like Google Dialogflow, Amazon Lex, and Microsoft Bot Famework/LUIS since circa 2018-2019.

It does feel like we're reinventing the wheel with some of this stuff, except letting the LLM do things that used to be handled with deterministic logic in an orchestration layer.

Given the current capabilities of the tech, I'd still prefer to do things the old way and only inject the LLM at the point in the workflow where I need to actually generate content, like for the email body. Rather than giving the LLM tools and telling it to "figure things out" and execute end to end.

The electoral college was created in the time before the Internet, computers, television, radio, telephone, telegraph, electricity, the automobile, the airplane, and the train. It was logistically impossible to have a national popular vote at the time. Even the gap between the election and inauguration was based on the time it would take a man on horseback to reach DC from the farthest point out in the country.

There was a highly publicized case a few years back where the police entered a hospital and ordered a nurse to draw a blood sample for an unconscious patient who had been in a car accident. They had no warrant and she refused per hospital policy (and law). The cops roughed her up pretty bad and arrested her.

https://www.npr.org/sections/thetwo-way/2017/11/01/561337106...

If Panther Lake and 18A are really on track for shipment later this year then why are they saying that its successor (Nova Lake) which is slated for 2H 2026 will use a mix of Intel and TSMC manufactured tiles?

If Panther Lake and 18A are really on track for shipment later this year then why are they saying that its successor (Nova Lake) which is slated for 2H 2026 will use a mix of Intel and TSMC manufactured tiles?

If only there were a way to make the models more efficient. Oh wait.

Tons of websites and books they scraped had copyright notices.

What they really destroyed was the idea that OpenAI would be able to charge $200/month for their ChatGPT Pro subscription which includes o1. That was always ridiculous IMO. The Free tier and $20/month Plus tier along with their API business (minus any future plan to charge a ridiculous amount for API access to o1) will be fine.

First of all, we need to just stop talking about AGI and Superintelligence. It's a total distraction from the actual value that has already been created by AI/ML over the years and will continue to be created.

That said, you have to distinguish between "good for the field of AI, the AI industry overall, and users of AI" from "good for a couple of companies that want to be the sole provider of SOTA models and extract maximum value from everyone else to drive their own equity valuations to the moon". Deepseek is positive for the former and negative for the latter.

In the long run (which in the AI world is probably ~1 year) this is very good for Nvidia, very good for the hyperscalers, and very good for anyone building AI applications.

The only thing it's not good for is the idea that OpenAI and/or Anthropic will eventually become profitable companies with market caps that exceed Apple's by orders of magnitude. Oh no, anyway.

Nope. Layoffs were always understood to be temporary, up through some point in the 1990s. Furloughs were much shorter in duration, typically days or weeks, and in some cases were partial (one or two days a week).

There was a time when the word "layoff" referred to a TEMPORARY separation due to a lack of demand with the understanding that when activity picked back up you'd be recalled back to work. This was particularly common in the automotive sector and really across manufacturing. These were cyclical industries and while employers couldn't afford to pay idle workers during periods of low economic demand, they also couldn't afford to lose the skillsets. Oftentimes unions would provide partial compensation to these workers until they were recalled.

Somewhere around the mid 1990s, "layoff" became just a euphemism for permanent reductions in force/downsizing.

>Maybe they really are all wrong

All? Quite a few of the best minds in the field, like Yann LeCun for example, have been adamant that 1) autoregressive LLMs are NOT the path to AGI and 2) that AGI is very likely NOT just a couple of years away.

My immediate reaction to the announcement was one of these is not like the others. OpenAI, a couple of big investment funds, Microsoft, Nvidia, and...............Oracle?

I do a lot of deep wilderness hiking. I have two Garmin products - an InReach Mini 2 and a Fenix 6X watch. There are a couple of big benefits over the Apple/Android ecosystem. 1) Battery life. 21 days (more with the newer models). 2) No dependence on cellular/wifi data availability (though they have been doing their damnedest to require it for syncing you watch to your Connect app). I'd love to see them merge the functionality of the two devices (provide satellite messaging direct from the watch). I have zero desire for some of the other stuff they've added (music player, payments, etc.), they're just battery drains.

The question seems malformed to me.

Text classification, clustering, named entity recognition, etc. are NLP tasks. LLMs can perform these tasks. ML models that are not LLMs (or even not deep learning models) can also perform these tasks. Is the author perhaps asking if the concept of a "completion" has replaced all of these tasks?

When I hear "traditional NLP" I think not of the above types of tasks but rather the methodology employed for performing them. For example, building a pipeline to do stemming/lemmatization, part of speech tagging, coreference resolution, etc. before the text gets fed to a classifier model. This was SOTA 10 years ago but I don't think many people are still doing it today.

>It's never made sense to me, but the standard explanation is that because they aren't accusing a person (the owner of the money), but only accusing an inanimate object

That is correct, but you need to understand the context. It originated in the 1600s as a way for maritime law to deal with pirate/smuggler ships who were operating in international waters, not flying the flag of any nation, and with no registered owner. Charging the ship and its contents with the crime rather than an unknown individual made sense in that context. Applying it to a car registered in the United States, driving down a highway in the United States, and being driven by a US citizen makes absolutely no sense because standard law can and should deal with that situation.

Hypothetically speaking, what would it mean if telescope technology were to improve to the point where we could see back to the point in time immediately after the CMB and what we saw were fully formed mature galaxies? It seems to me that there would only be a couple of possible explanations - either the CMB would have to be re-interpreted as some sort of local (but extremely large scale) event not necessarily from the formation of the entire universe or else it would be evidence of a Big Bounce where somehow parts of a previous universe survived.

>Eventually, you’ll just need to connect a model to a computer to solve most problems - no complex engineering required.

The word "eventually" is doing a lot of work here. Yes, it's true in the abstract, but over what time horizon? We have to build products to solve today's problems with today's technology, not wait for the generalized model that can do everything but may be decades away.

>"The two main categories I see are people who think AI agents are obviously things that go and act on your behalf—the travel agent model—and people who think in terms of LLMs that have been given access to tools which they can run in a loop as part of solving a problem."

This is exactly the problem and these two categories nicely sum up the source of the confusion.

I consider myself in the former camp. The AI needs to determine my intent (book a flight) which is a classification problem, extract out the relevant information (travel date, return date, origin city, destination city, preferred airline) which is a Named Entity Recognition problem, and then call the appropriate API and pass this information as the parameters (tool usage). I'm asking the agent to perform an action on my behalf, and then it's taking my natural language and going from there. The overall workflow is deterministic, but there are elements within it that require some probabilistic reasoning.

Unfortunately, the second camp seems to be winning the day. Creating unrealistic expectations of what can be accomplished by current day LLMs running in a loop while simultaneously providing toy examples of it.

This whole idea of prompting an LLM and piping the output as the input (prompt) of another LLM and asking it to do something with it (like critique/edit it) and then piping the output of that LLM back to the first LLM along with instructions to keep repeating the process until some stop criteria is met seems to me to just be a money-making scheme to drive up token consumption.

The gap between the claimed capability of "AI Agents" and what you're actually able to build with tools like AutoGen and Crew (and presumably AgentForce, it's been a few months since I saw a demo) is the largest gap I have ever seen in the field and I've been working with NLP/Conversational AI professionally since 2016.

I'm glad to see someone speaking truth on this topic. There are lots of interesting things happening in quantum computing RESEARCH but it is all at an experimental stage. I liken it to the early days of experimentation with vacuum tubes in the history of classical computers.

Ok, I just tried again.

Turned off WiFi and put phone into Airplane Mode. Verified that Bluetooth is still connected to the watch.

Go into the Connect app. There's a big red box that says "No Internet Connection."

Heart Rate is transmitting to the Connect App in real time.

However, if I try to run a Sync from Connect, it hangs and then fails.

Turn on WiFi and take phone out of Airplane Mode. Run a Sync from Connect. It completes successfully.

Confirmed - If you complete an Activity and want to sync your watch to your phone with the Connect app, your phone must have an Internet connection.

Are you sure about that? I have a Garmin Fenix 6 and the Connect app on Android. While it's true that all data is stored locally on the watch and it doesn't need to be tethered to a phone to work, I'm pretty sure that syncing the watch to the Connect app does require the Connect app to be connected to the server. I have experienced this multiple times, after completing a hike in a remote area and I try to sync the Hike Activity to Connect it fails due to not having a cellular connection despite the watch being connected to the phone via Bluetooth.

A lot of modern AA missiles don't use any traditional shrapnel at all. They use what's called a continuous rod warhead. It's a bunch of steel rods arranged as a cylinder with alternating ends welded to each other on each side effectively making it one long rod. The explosive charge is placed in the middle and when it goes off the rod expands into one big rod in a circle which then cuts the aircraft in half.

Train a model on a data set that has had all instances of small talk to close a conversation stripped out and see if the models evolve to add closing salutations.

>I doubt we will talk a lot about AI in five years because AI will be an integral part of how people search for and interact with information. Or, if we talk about AI in five years, it will most likely refer to different technologies and solutions than what we denote as AI in 2024.

It's the latter. In 5 years, what we consider to be AI today will just be standard product features and the AI being discussed will be some new set of capabilities that are just being developed, and the same will happen 5 years after that.