DebtDeflation

Simple. "AI, in its present incarnation, is not capable of doing it all. So let's instead focus on where specifically we can apply it to derive some benefits."

> most "AI Agents" that make it to production aren't actually that agentic. The best ones are mostly just well-engineered software with LLMs sprinkled in at key points

I've been saying that forever, and I think that anyone who actually implements AI in an enterprise context has come to the same conclusion. Using the Anthropic vernacular, AI "workflows" are the solution 90% of the time and AI "agents" maybe 10%. But everyone wants the shiny new object on their CV and the LLM vendors want to bias the market in that direction because running LLMs in a loop drives token consumption through the roof.

Unfortunately, that ship sailed a long time ago. Why is no one in the administration paying any attention to the outsourcing of high skill knowledge work to India and elsewhere? Obviously I have a bias working in technology, but it seems to me to be a much more CURRENT issue and one that can actually be addressed in the present.

That's a fair assessment. If they're not shipping 18A in volume early next year they likely will end up getting parted out later next year.

Beyond ensuring adequate cash flow, they need to be 100% focused on getting 18A shipping in volume as soon as possible rather than financial engineering stuff.

A company with $350B in annual Revenue, $200B in Gross Profit, $100B in Cash in the bank lays off a few hundred people making a few hundred thousand dollars a year each, the savings amount to rounding errors. It's hard to understand what these Tech companies are trying to accomplish with the never ending rolling layoffs other than to engender fear in the workforce. It's like all that Great Resignation and Quiet Quitting stuff a few years back broke the collective brains of the C-Suite and they decided they just cannot tolerate labor having any sort of bargaining power at all and needed to mete out punishment to put workers back in their place.

They use Customer Service as their domain example. This is an area that I've spent the last decade applying AI/NLP to. 99% of tasks in this domain are fully deterministic - get information about a product, place an order, cancel an order, get status on an order, process a return, troubleshoot a product, etc. These are all well-defined processes that should make use of orchestrated workflows that inject AI at the appropriate time (to determine the customer's intent using a classifier, to obtain information for slot-filling using NER, and lately to return information from a knowledgebase using RAG). Once we know what the customer wants to do and the information required, we execute the workflow. There's no reason to use an "autonomous agent" to engage in reasoning and planning. Unless we just want to drive up token costs for some reason.

His position seems to be that Quantum Mechanics is a statistical description of what we observe not the underlying theory of what's actually happening. He also believes that superposition is a mathematical trick and not real.

The factories of today are nothing like the 1980s factories this administration thinks they are going to bring back. Go to FRED or any other economic data site and plot manufacturing output and manufacturing employment on the same chart. The US manufactures more today than we ever have in history and it grows every year. Meanwhile, manufacturing employment is around its all time low and declines every year.

In the Task Automation demo, how does it know all of the attributes of the motorcycle he is trying to sell? Is it relying on the underlying LLM's embedded knowledge? But then how would it know the price and mileage? Is there some underlying document not referenced in the demo? Because that information is not in the prompt.

The issue is the scale of the improvements. GPT-3.5 Instruct was an utterly massive leap over everything that came before it. GPT-4 was a very big jump over that. Everything since has seemed incremental. Yes we got multimodal but that was part of GPT-4, they just didn't release it initially, and up until very recently it mostly handed off to another model. Yes we got reasoning models, but people had been using CoT for awhile so it was just a matter of time before RL got used to train it into models. Witness the continual delays of GPT-5 and the back and forth on whether it will be its own model or just a router model that picks the best existing model to hand a prompt off to.

Most large companies monitor devices. You start GMailing source code to yourself, you get fired.

> Since trade is conducted largely in USD, that means other governments must purchase USD to trade.

This is just one facet of a broader point. When people talk about a "trade deficit" what they are really talking about is what's known in Economics as a "Current Account deficit". When discussing international trade and Balance of Payments, there are two accounts - Current Account (goods and services flow) and Capital Account (asset and liability flow). By definition, the two must net to zero. That's not an equation, it's an identity. If you have a Current Account deficit then you have an equivalent Capital Account surplus. It works in both directions. For example, foreign direct investment into another country in the form of a loan leads to a flow of funds into the country (Capital Account surplus) and then those funds are used to buy things (import) from the first country or other countries, leading to a Current Account deficit.

And the LangChain definition is a further simplification of Anthropic's:

"An AI agent is a system that uses an LLM to decide the control flow of an application."

What gets left unsaid is whether current SOTA LLMs actually have the reasoning and planning capabilities to do this reliably. I would argue that except for code debugging tasks and simple research tasks (iterative Googling with a web browser and then formatting the results into a report format) they do not. That may change in 6 months, but right now the hype has gotten ahead of the capability.

In the United States, if you buy "Denatured Alcohol" at a big box store, it will be a toxic brew of often 50-60% methanol as well as smaller amounts of MEK, MIBK, and even acetone in addition to the ethanol. However, you can order "Specially Denatured Alcohol (SDA)" from various chemical suppliers or even Amazon. "SDA 40B 200" will be 99.5% ethanol with 0.1% tert-butyl alcohol and 0.0005% denatonium benzoate. It's what I use for camp stove fuel since I refuse to carry methanol in my pack with food.

What if they considered that and determined that at typical consumer SaaS pricing of low tens of dollars per user per month it's "still impossible to make money"? What if they went a step further and looked at typical Enterprise SaaS pricing (low hundreds of dollars per seat per month) and determined "still can't make money"?

Shortly after ChatGPT hit the scene, everyone said "Google invented this technology, how could they fall so far behind in commercializing it, haha they're IBM now".

Maybe they didn't fall behind in anything, maybe they just did an analysis of what it would cost to train transformer models with hundreds of billions of parameters, to run inferencing on them, and then decided that there was no way to actually be profitable doing this.

I've been building workflows with "AI" capability inserted where appropriate since 2016. Mostly customer service chatbots.

99.9% of real world enterprise AI use cases today are for workflows not agents.

However, "agents" are being pushed because the industry needs a next big thing to keep the investment funding flowing in.

The problem is that even the best reasoning models available today don't have the actual reasoning and planning capability needed to build truly autonomous agents. They might in a year. Or they might not.

Fun fact: In Texas, ALL traffic offenses (except for speeding and open container) are considered arrestable offenses. It's up to the discretion of the officer whether he writes you a ticket or hauls you off to jail.

There are two competing definitions of agents being used in industry.

https://www.anthropic.com/engineering/building-effective-age...

"- Workflows are systems where LLMs and tools are orchestrated through predefined code paths.

- Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks."

What Anthropic calls a "workflow" in the above definition is what most of the big enterprise software companies (Salesforce, ServiceNow, Workday, SAP, etc.) are building and calling AI Agents.

What Anthropic calls an "agent" in the above definition is what AI Researchers mean by the term. It's also something that mainly exists in their labs. Real world examples are fairly primitive right now, mainly stuff like Deep Research. That will change over time, but right now the hype far exceeds the reality.

I once saw the Heisenberg Uncertainty Principle described as the result of the simulation creators using an 8 bit variable to store position and momentum combinations which were each themselves 8 bit values.

> an agent has to perform the targeted tasks internally: they "dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks".

> What most agent startups are currently building is not agents, it's workflows, that is "systems where LLMs and tools are orchestrated through predefined code paths." Workflows may still bring some value

While this viewpoint will likely prove correct in the long run, we are pretty far away from that. Most value in an Enterprise context over the next 3-5 years will come from embedding AI into existing workflows using orchestration techniques, not from fully autonomous agents doing everything end to end "internally".

This is known as the "Hygiene Hypothesis".

https://en.wikipedia.org/wiki/Hygiene_hypothesis

The basic rationale for it is that the human genome did not evolve in a sterile environment, it evolved in an extremely hostile environment, and so developed pretty harsh defense mechanisms, which if not trained (via early exposure) on the appropriate targets will instead learn to target harmless compounds as well as the body itself.

Even moreso, it only applies to initial model training by companies like OpenAI not other companies using those models to generate synthetic data to train their own models.

Totally agree. Even the 3.5 years in the role that Pat Gelsinger got was insufficient to see his strategies come to fruition. At this point it's baked and there's not much anyone can do. My point isn't to assign credit/blame, it's simply to point out that Intel will not survive in its current form if they are not shipping 18A products by this time next year.

It's really simple - either Intel is shipping products on 18A (Panther Lake and Clearwater Forest) by Q1 of next year or they are not, and their entire future hinges on this.

I saw Musk saying a couple of days ago that we've "hit the limit of peak data" for training AI. My immediate reaction was no, surely you have not trained on every copyrighted textbook on every subject ever written. You hit the peak of easily accessible internet data that you could quickly steal to train your models.

That's what I'm trying to understand. His educational background was in Physics/Nuclear Engineering so he's obviously a smart guy, and he was CEO/Chairman of Cadence for 15 years, but other than that his 40+ year career has most been in VC and being on the boards of an incredibly large number of companies.

https://en.wikipedia.org/wiki/Lip-Bu_Tan

https://www.linkedin.com/in/lip-bu-tan-284a7846/details/expe...

It's a fair point. And even short of "going to zero" the Nikkei didn't recover its 1989 high until last year. 35 years is a long time to get back to breakeven.

SP500 is down like 8% from the ATH, which BTW was less than 3 weeks ago. People need a little perspective. I lived through the GFC and the Dotcom bust, this is nothing (so far).