Jul

13

The Turing Test, originally introduced by British mathematician and computer scientist Alan Turing in his 1950 paper "Computing Machinery and Intelligence," is a proposal for evaluating whether a machine can exhibit intelligent behavior indistinguishable from that of a human.

This is more relevant today than ever. Applied to the market some moves appear very machine like. Other moves not so much. A large proportion if trades now are computer generated. Maekt making must all be computer managed. Humans still can game the machines as their rules are fixed.

Peter Ringel writes:

The Turing Test is fuzzy. Useful, but not precise. The underlying question is: Is AI conscious? My answer is: NO! but it is a freakishly good imposter.

It is also domain specific. In my opinion, in the general domain, AI broke the Turing Test in 2024. Humans can't distinguish anymore responses from robots to humans in the general domain. In specific domains it is different. The low to mid level is broken, the mid to high is not. OTOH, recently we had news that open AI proofed some longstanding Math hypothesis (I lack the knowledge to judge). And we all know about Google's AlphaFold. Both domain specific. High End.

Market making was algo based before AI. I indeed, try to game this with my human discretion. (At least in the sort-term part of my trading. It is not relevant for longer TF.) AI definitely facilitates research. This makes markets more efficient and faster efficient. I think we are seeing this currently.

Creativity is not broken. I hope, as of now.

Oliver Joseph writes:

A conversation that I find endlessly fascinating. In my opinion the current technology stack with regards to LLMs and machine learning tools in general is an absolute game changer. The barriers of entry associated with software seem to have been lowered significantly, however to use another tech analogy the "last mile" still remains elusive. By that I mean that the LLM really needs help to get certain complex problems over the finish line. That said, what we are dealing with is essentially a cliche generator. Humans tend to anthropomorphize, and personally I have found myself pondering the meaning of consciousness as I use the latest generation of LLM's ( I don't think we are dealing with consciousness here). Rich Suttons essay "The Bitter Lesson" from 2019 concludes with the idea that we want "AI agents that can discover like we can not which contain what we have discovered" arguing that by using experts to build in their knowledge it makes it harder to even see how the discovering process happens in the first place. He emphasizes a focus on what he calls scaleable general purpose methods primarily search and learning. It is interesting to me as I work with locally run models to see what output the <| think |> control token actually generates. I have one of those old IBM notepads that says think on it, however I must say I am prodded to think myself as I see this machine chugging through the art of thinking clearly with such mechanical precision (now we're in Dobelli of the beast). In conclusion I agree with the above sentiment that creativity is not broken but rather like where T.S. Eliot was wrestling with the concept of historical sense in his art; we now find ourselves with a large portion of human knowledge existing simultaneously in one place frozen in time, embalmed and pacified. It is up to us to be creative and let the tool inform our creative throughput.

On another tangent when I produce and or record music my workflow (and output) is changed by the tools I use. Ableton on a Macbook with VSTs gives me a different feeling and results than using an Atari 1040st and an Akai S950. I get the same feeling as I use LLM's for certain tasks.

Zubin Al Genubi adds:

I find AI helpful but AI doesn't quite get the joke. Kind of like a super-smart nerdy guy.

Nils Poertner comments:

From a spec point of view, there are tremendous opportunities where human creativity is always finding something which LLMs not seeing…as they have built mirrors in a mirror rooms and confusing this room with reality (but reality remains mysterious, indescribable and so on…and fuzzy logic or jain logic is already a slight improvement to our "either-or" logic that is so prevalent in our times).

Peter Ringel responds:

Yes, the tools we use define culture. Music culture, trading culture - binned per generation. AI is a tool (as of now).

If I close my eyes, I still can hear the metronome signal of Cubase through a crappy 1040 sound speaker ( or was it the monitor ?): "meeeep, mup mup mup" for a four to the floor. Cubase and similar tools made it easy to create a typical track structure. Probably long and repetitive. This defined the 90 and 2000 pop music. Before that - with some overlap - Sound-tracker. The UK in the 90 is unthinkable without it. Way more interesting. Then Fruity loops and similar. More questionable structure. Then Ableton. It freed the structure again, the creativity. I don't know, what the kids prefer now.

Nils Poertner responds:

Yes, very good. Real speculators are incredibly good in finding opportunities in mkts (and not just go with the latest fashion and think that's it). For older specs it will be "how to keep and possibly improve health" and for younger ones, I suppose it is how to turn their frustration/ anger into creativity and learn the game.

Oliver Joseph writes:

I love that Peter. Yes, Cubase was and is a fantastic tool precisely because of its constraints.

LLM’s being another cool new tool it is worthwhile to make an attempt to understand its uses and constraints. As a very simple thought experiment I asked three different models de jour to generate a random number between 0 and 500. I ran 500 iterations of the prompt per model. The “think” tokens generated reasoning from the models that stated basically “I’ll just pick a number that sounds random”. The “think” token of course being obscured from the end user. Going through such gesticulations is completely unnecessary other than I think that conceptually it's important to understand that here we ask for an output (the random number) and what we actually get is a number within the bounds of the below graph. Importantly we run the risk of thinking it's what we actually asked for. Surprise! It's not. Everyone on the spec list knows the right tool for the above job is a HRNG or TRNG or who knows what all else; lava lamps and wristwatches.

To my mind it is one of those inexplicably beautiful facts of history that Wall Street was thus named for a wall that separated traders and speculators from livestock. In the car business we have a large vocabulary derived mostly from farming and the cattle business. One thing that we would commonly do is go out and “put hands on” our “stock” go physically touch and look at the car. Statements are a representation of inventory I am supposed to have and obviously they have their use.

In summary as has been the case for the last 6000 years of recorded human history the farmer has asked the farmhand “where is the cow?” and the farmhand sometimes just said “over by the watering hole.” even though he actually had no clue. We equate the representation of the thing and the thing in and of itself at our peril. As speculators we should endeavor to identify the places in our experience where an incorrect representation is believed in more so than the thing in and of itself. I believe it pays to every now and again to hop the liminal.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search