Mustafa Abualsaud

Building high quality retrieval systems. Search (IR)/AI/LLMs. Ex-Bloomberg AI. PhD from @UWaterloo.

[Mustafa Abualsaud @ammsa]

@lejeunesimon @pixelzso Yes.

[DAN KOE @thedankoe]

When calories became cheap, convenient, and abundant, bodies decayed.

Thinking is now cheap.

[Naval @naval]

Don’t send me the report, just send me the prompt.

[Sahil Bloom @SahilBloom]

I’m convinced that you can get pretty far in life by just finishing things. The world is full of half-written books, half-built businesses, and half-kept promises. You stand out by showing up. By closing loops. By doing what you said you’d do. By having the courage to finish.

[Alexis Ohanian 🗽 @alexisohanian]

Very well articulated. Taste is going to keep compounding in value as it gets easier and easier to create

[Mustafa Abualsaud @ammsa]

@GrubesicProfits @Gubloinvestor What makes questrade better than WS?

[Mustafa Abualsaud @ammsa]

@optionscjp They have a feature where they show catalysts behind notable price movements. Earning calls summaries and transcripts are nice additions.

[Jimmy Lin @lintool]

Thus, our conclusions: This I believe is the first demonstration of the need for hybrid search. Hence the claim that hybrid search is a @UWaterloo innovation. You're welcome!

The broader lesson is that old baselines are still surprisingly important. Let's not forget them.

[Andrej Karpathy @karpathy]

This is the the quote I've been citing a lot recently.

[Mustafa Abualsaud @ammsa]

@RahafsView The best is still ahead

[Mustafa Abualsaud @ammsa]

@RahafsView $AMD

[Haider. @haider1]

Google Jeff Dean says bigger context windows alone are not enough

What matters is staged retrieval: lightweight mechanisms that narrow a trillion tokens down to 10 million, then to the million you actually need

"you don't need a trillion at once, you need the right million"

[Aaron Levie @levie]

It’s wild to think about what types of infrastructure and services must change in a world where agents can process information a hundred or a thousand times faster than humans.

Even the tools that were built for machine speed before, generally were still in service of end-users making a request somewhere in the system. Agents running 24/7 and in parallel modify these requirements meaningfully. Here are just a few examples:

* Sandboxes. Agents need sandboxes to operate in that have to be insanely low latency because they can boot up these environments for coding at any moment.

* Search (both publicly and within an enterprise). Agents can parallelize searches hundreds or thousands of times so they need to be able to work with fast indexes of information.

* Payments. Agents can now pay in micro transactions, and aren’t bothered by the friction of paying $0.01 for a resource that a human would be.

* File systems. Agents need to be able to work with files at a scale that humans never had to worry about. You’ll have all new complexity around version control, permissions, and having agents reading/writing from data at insane speeds.

And there are tons more. We’re going from a word where software was built for people to a world where it’s built for agents. Lots of changes downstream as a result.

[Mustafa. @oprydai]

i often think about this..

[Leonie @helloiamleonie]

Context engineering is probably 80% agentic search.

[Anush Elangovan @AnushElangovan]

100%.

I have seen domain experts generate more code in the last month than their entire lives.

[Jo Kristian Bergum @jobergum]

Many of you know me as the bm25 guy but I'm afraid there is no single silver bullet for solving retrieval for agents. You also need to make the data retrievable in the first place.

Yes, agents are great at formulating queries and they are relentless so classic issues like vocabulary missmatch are arguably less of a concern when the agent is doing the queries. But, a lot of the context that we would like to feed agents with it isn't represented in text so embeddings are still very important to cover more modalities. Making more data useful.

We now see the early raise of multi-modal embeddings handling a mix of modalities into the same embedding space which I'm directionally excited about.

For example, there is a lot of talk about the context layer for enterprises and especially here multimodal representations is key. Think meeting nodes, audio, images (e.g images of pdf pages).

So even if I'm the bm25 guy, I still think embeddings will play a big role in the context layer for agents.

[Mustafa Abualsaud @ammsa]

@RealMattMoney Yes. Mobile UX needs work. Navigation feels clunky and it’s not optimized for mobile screens.

[Mustafa Abualsaud @ammsa]

@softwaredoug https://t.co/hcDuD1MR1N

[Gergely Orosz @GergelyOrosz]

"I don't like pull requests (PRs) any more. A large chunk code change doesn't tell me much about the intent or why it was done.

I now prefer prompt requests. Just share the prompt you ran / want to run.

If I think it's good, I'll run it myself and merge it." - @steipete

wow

[Mustafa Abualsaud @ammsa]

@GergelyOrosz Strong academics + serious research + hackathons culture, and the co-op pipeline turns that into real-world skills. It’s also not exactly a party town, so there is not much to do but grind, build, and get really good.

[Gergely Orosz @GergelyOrosz]

Inside a group of ~50 CTOs + founders. Talking about hiring new grads. 4 of them violently agreeing to how amazing University of Waterloo [Canada] is doing something so right - and their goto place to hire standout interns + new grads.

What is Waterloo doing others are not?