zwaps

This reads like the handbook for people making grind-based games. Sure enough, the author exclusively works in the mmorpg space.

If you are a game designer, please take this with a grain of salt.

Fun does not equal repeated challenges. And let me also reject the implicit notion that stories are entertainment but not, academically speaking, fun.

Do people find this sort of writing appealing?

For me, it's cringe to borderline painful to read.

The OpenAI numbers are a red herring anyway.

For most plans, Deep Research is capped at around 20 sources, making it for many cases the least useful research agent, in particular worse than a thinking mode Gpt5 query

Yes you are. I can show you 15 line SQL queries you would be hard pressed to understand, let alone built with precision as a novice

And you want everyone to learn this? People who don’t even have time or ability to master excel?

Really?

Its laggy on mobile

Scrolling initially I thought: Design of what?

On mobile, the showcase is also not really visible.

Define neoclassical economics beyond its use in the blogossphere

The reaction is utterly sound. What issue do you have with these very clearly laid out and logical reasons?

Interesting in the second graph, the countries which we traditionally consider to be well working societies are those where out-hate has not crossed in-love

Interesting to see this now launching, when most companies have their own customGPT solution and MCP makes headway towards decoupling the frontend layer. Data seems to be stored outside of the customers control, so this will be a difficult sell for many companies.

What type of businesses are you targeting?

There is a career book that makes a lot of good but extreme points.

One of them is: technical ability is actively detrimental to your power and career. You have to spend time and energy on actually doing things, and every competent manager will do their best to keep you right where you are, with as little political influence as possible.

Conversely, as a manager, so the book says, you want to avoid actually doing anything. You should start initiatives -as many as you can- and deftly use your political capital to either own, disown, or weaponize them. Whether they succeed in creating value is irrelevant, certainly not something you should focus on.

People focused on success and value of initiatives are still working hard when you have moved on. These people are hopelessly behind the scheming manager, eating crumbs.

And if necessary, you the manager just claim credit retroactively.

I am so tired of these undifferentiated takes.

These types of articles regularly come from people who don't actually build SCALE systems with LLMs. Or, people who want to sell you on a new tech. And the frustrating thing is: They ain't even wrong.

Top-K RAG via vector search is not a sufficient solution. It never really was for most interesting use-cases.

Of course, take easiest and most structured - in a sense perfectly indexed - data (code repos) and claim that "RAG is dead". Again. Now try this with billions of unstructured tokens where the LLM really needs to do something with the entire context (like, confirm that something is NOT in the documents), where even the best LLM loses context coherence after like 64k tokens for complex tasks. Good luck!

The truth is: Whether its Agentic RAG, Graph RAG, or a combination of these with ye olde top-k RAG - it's still RAG. You are going to Retrieve, and then you are going to use a system of LLM agents to generate stuff with it. You may now be able to do the first step smarter. It's still Rag tho.

The latest Antrophic whoopsy showed that they also haven't solved the context rot issue. Yes you can get a 1M context scaled version of Claude, but then the small/detail scale performance is so garbage that misrouted customers loose their effin mind.

"My LLM is just gonna ripgrep through millions of technical doc pdfs identified only via undecipherable number-based filenames and inconsistent folder structures"

lol, and also, lmao

Yes but that would be worse than many RAG approaches, which were implemented precisely because there is no good way to cleanly search through a knowledge base for a million different reasons.

At that point, you are just doing Agentic RAG, or even just Query Review + RAG.

I mean, yeah, agentic RAG is the future. It's still RAG though.

Wait, this isn’t normal?

Why are Dutch or Swiss highways so much more pristine than French and German ones?

Have you compared distances?

Why do you have to keep the dataset in memory? We had good distributed streaming datasets for a good while now, no?

In recent weeks, doing a deep research with gpt5 seems broken for me

It usually looks at less than 20 sources. Basically, it has become useless.

Conversely, just doing thinking or pro mode gives me hundreds of sources

What gives?

At this point there's about a hundred (or more) years of research, experience, law etc. in competition policy.

Competition doesn't necessarily just win in the mid-term. Competitions works - if the conditions are right, markets are anonymous and dynamic, not two-sided, not plaqued by information symmetries and - importantly - vertical entanglements. Competitions also works out in the long term. Like, take a hundred years and squint.

However, in the meantime of all this, there are many cases where the market outcome moves strikingly far away from the optimum. What that means is that the market situation destroys value (consumer welfare, societal roi, whatever)

You can scan the OP for about three sections and see that Google is violating any reasonable and established take on how market regulation, leading to an inefficient market outcome.

This is not some special European temperament. This is just standard and - just to make this clear - 100% American economic theory as previously applied and pioneered mainly in the good US of A. If this doesn't get applied in the US now, we may call this regulatory capture.

Personally, I feel it also really speaks to the situation that Google is lauded as representing the US tech scene. I disagree here. I think the US tech scene goes far beyond Google. Google ain't even a particular strength, probably more of a weakness by now.

By contrast, you could (and should) bring up about a million things the EU and the commision in particular does to stifle a EU tech scene. Bog standard application of competition policy ain't it.

How does it look with ColBert style late interaction embeddings?

No one means Matryoshka embeddings when they talk about sparse embeddings. This is not pedantic.

if you manage to do this on ios I will give you money

I have applied O3 pro on unpublished abandoned research of mine that was never published and lives in an intersection that is as entirely novel as it's uninteresting.

O3 pro (but not O3) was successfully able to apply reasoning and math to this domain in interesting ways, much like an expert researcher in these areas would.

Again, the field and the problem is with 100% certainty OOD of the data.

However, the techniques and reasoning methods are of course learned from data. But that's the point, right?

You actually do not classify digits that didn't appear, you classify different pictures of digits that DID appear.

Similarly, LLMs do not invent a new way of reasoning about problems or language. They do, however, apply these to unseen problems.

LLMs are one level of abstraction up, but it's a very interesting level of abstraction.

Scale changes the performance of LLMs.

Sometimes, we go so far as to say there is "emergence" of qualitative differences. But really, this is not necessary (and not proven to actually occur).

What is true is that the performance of LLMs at OOD tasks changes with scale.

So no, it's not the same as solving a math problem.

The author seems to fundamentally misunderstand how MCPs are going to be used and deployed.

This is really obvious when they talk about tracing and monitoring, which seem to be the main points of criticism anyway.

They bemoan that they cant trace across MCP calls, assuming somehow there would be a person administering all the MCPs. Of course each system has tracing in whatever fashion fits its system. They are just not the same system, nor owned by the same people let alone companies.

Same as monitoring cost. Oh, you can’t know who racked up the LLM costs? Well of course you can, these systems are already in place and there are a million of ways to do this. It has nothing to do with MCP.

Reading this, I think its rather a blessing to start fresh and without the learnings of 40 years of failed protocols or whatever

I get the client part, great job!

However, the agent is really just a wrapper of Langchain AgentExecutor. This doesn't seem like something someone would want to put into production.

Wasn't this already implemented in some agents?

I want to remember I heard about it in several podcasts

Gemini loses coherence and reasoning ability well before the chat hits the context limitations, and according to this report, it is the best model on several dimensions.

Long story short: Context engineering is still king, RAG is not dead

Very cool results, very comprehensive article, many insights!

Media literacy disclaimer: Chroma is a vectorDB company.

It is true though. Like, literally. Why do you think it is called Lite?