Joe Fitzsimons

Physicist. Founder and CEO @horizon_quantum. Former @quantumlah PI and associate professor. Slowly drifting south-east (Dublin-Oxford-Singapore).

[Joe Fitzsimons @jfitzsimons]

Delighted to be able to share that we expect to begin trading on Nasdaq tomorrow as $HQ. After 7 years of work, it’s great to finally reach the starting line.

$DMYY

https://t.co/9tJ8iMXemJ

[Joe Fitzsimons @jfitzsimons]

@DesFrontierTech We want to get to the point where you don’t need to think about quantum at all, where you can just have seamless acceleration. But we have quite a bit to do still if we are to reach that point.

[Joe Fitzsimons @jfitzsimons]

What we have been working towards is the ability to automatically accelerate programs written in classical languages. The techniques we have developed in that direction naturally lead to hybrid programs with multiple calls to quantum processors. Automating the choice of what goes where in that context is certainly something we have thought about.

[Joe Fitzsimons @jfitzsimons]

@DesFrontierTech Absolutely, although it does not currently do this. Part of the reason we explore characterisation and error mitigation in this way is to find ways of eeking better performance out of quantum processors, to try to pull forward quantum advantage.

[Joe Fitzsimons @jfitzsimons]

We’ve had some nice results multiplexing characterisation experiments. Our latest blog post digs into our team’s work on ZZ and control crosstalk characterization protocols that scale logarithmically in the number of qubits in a processor.

[Horizon Quantum @horizon_quantum]

Maximizing hardware performance is key to reaching broad quantum advantage, making efficient, scalable characterization increasingly important. Our proprietary multiplexed protocol can enable rapid characterization of high-qubit-count systems. Learn more 👇https://t.co/uaQyAqmCb2

[Joe Fitzsimons @jfitzsimons]

@QuantumDevIO @horizon_quantum It’s nothing to do with the multiplexing. On some qubits readout error or short coherence time means that you don’t see oscillations clearly at the time increments we chose whether they were multiplexed or not.

[Joe Fitzsimons @jfitzsimons]

@appenz Oracles aren’t real life.

And even if they were, “broad set of tasks” is kind of misleading. AI is also -provably- useless for a broad set of tasks because random functions almost always have no structure. But people only care about a set of measure zero that is atypical.

[Joe Fitzsimons @jfitzsimons]

@BullMeechum3 @quantumashley @ORNL @fzj_jsc QUBO is polynomially equivalent to weighted Max-2-SAT, so every decision problem in NP can be reduced to QUBO in polynomial time. So it’s not just lattices and you don’t need Mythos, it’s pretty much all of cryptography. But, we don’t know of an efficient algorithm for all QUBO.

[Joe Fitzsimons @jfitzsimons]

@quantum_aram @henryquantum @AAmbainis That comment seems to make it more complicated than it actually is, since you can get a subexponential scaling in the number of gates in general, rather than requiring it to be quasilinear in the number of qubits. It seems like they parameterised it in an awkward way.

[Joe Fitzsimons @jfitzsimons]

@QuantumDevIO @horizon_quantum We already have detailed resource counts at the hardware, hydrogen and helium levels, so you can see how resources change as code becomes more hardware specific (since changing instruction set, embedding in the connectivity graph etc have a direct bearing on resource usage).

[Joe Fitzsimons @jfitzsimons]

We have signed a letter of intent to acquire Entropica Labs. For more details see the press release here: https://t.co/stXTSO2PlB

[Joe Fitzsimons @jfitzsimons]

TypeSafe AI seem to talk quite a bit about the length of their development of Jev, so it’s been fun to see Blatent Struct come together over a much more compressed timeframe with much broader capabilities.

[Joe Fitzsimons @jfitzsimons]

Rob was a visiting student in my research group and my brother is his co-founder, so you can take what I say with a pinch of salt, but LLMs that are 10x faster and 100x cheaper with the output guaranteed to be in exactly the form you need seem a pretty compelling proposition.

[Robert Pisarczyk @R_Pisarczyk]

Very excited to finally share what we've been building!

Today, we are launching Blatent Struct, our first language model. 🚀

One thing that's always bothered us about building with LLMs is how much engineering you need around the model. Getting an answer is one thing. Making sure it follows the right decision path, returns everything your application needs and doesn't break your business rules is another.

We wanted to change that.

Struct brings programmability to language models.

You can define complex, cascading decisions, conditional dependencies and exactly what each outcome must satisfy. You can also enforce business rules and permissions during inference.

And rather than stitching together multiple model calls, Struct can handle the whole decision tree in a single run.

We also spent a lot of time making sure this doesn't come at the expense of performance.

In our benchmarks, Struct achieves comparable accuracy to Claude Sonnet 5.5, at 11× the speed and over 100× lower estimated cost.

It's also over 2× faster than Jev, with higher average accuracy across 20 benchmarks. And on source-grounded extraction, it outperforms Cloudflare Clairic (87.5% vs. 81.7% across six benchmarks), at 62% lower estimated cost.

For me, the exciting part is what this enables: everything from document processing and customer support to agents that can make complex decisions while respecting the rules you've set.

This is just our first model, and there's a lot more to come.

I'd love for you to give it a try and tell us what you think. We're especially keen to hear what works, what doesn't and what you'd like us to build next.

🚀 Get started for free (up to 100m input tokens are on us) : https://t.co/iwEo6Cybo2

[Joe Fitzsimons @jfitzsimons]

@nick_farina @letonyo You know what are capable of recursive self improvement? Humans.

[Joe Fitzsimons @jfitzsimons]

@letonyo I find it really odd. Humanity discovers that a solving a bunch of open maths problems can be reduced to a finite sequence of matrix operations and people are mourning this rather than celebrating it.

[Joe Fitzsimons @jfitzsimons]

@nick_farina @letonyo To be honest, I’m very excited about what having this new tool will enable in physics and mathematics. If I was starting as a PhD student today, I’d definitely focus on harnessing agents as part of the research process. I think there is potential for a new golden age of discovery

[Joe Fitzsimons @jfitzsimons]

@nick_farina @letonyo I don’t think it will be. We just broke through a barrier so progress is happening so fast it looks like there is no limit, but there will be new limits. We just don’t know what they are yet.

[Joe Fitzsimons @jfitzsimons]

@cw_3245 @letonyo In my opinion, there is zero chance this kills mathematics as a field. It just changes what’s easy. But what it does mean, I would think, is that there are now tremendous opportunities to harness recent AI advances to push the frontier of mathematics.

[Joe Fitzsimons @jfitzsimons]

Close, ChatGPT. Close.