417 results · Most relevant
There is no clear-cut answer for how many harnesses to build; consider the spectrum from embedding everything into skills to building from the ground up, and ask how generic or specific a harness should be and how much effort to invest.
described as open question
Francesco notes that if a decision — like the 20K lending on the same day as payment three — influences the final decision, it definitely comes with accountability.
When a ballpark total doesn't feel right, finance experts react instinctively, e.g. 'how did we spend this much on marketing, I don't think that's right'.
described behavior
The harness instructions are basically instructions for instructions — the IKEA guide — supposed to guide you nicely on the things you should care about, but then you can pick your preferences on how they get implemented.
Anekha describes using alternative data like satellite information and credit card information on the buy side, and states her number one question was whether vendors were actually looking at her data.
From data engineering experience, metric definitions and dimension definitions always change. A common data conversation is 'is this number correct? I don't know. It sounds like it is, but we're never sure' — and if we're never sure, is the number correct?
Anekha Sokhal describes conference dynamics: after her talk, people messaged her on LinkedIn within the past hour asking for meetings in Houston or New York, and the attendees included traders at funds and consulting firms. The speakers also endorse the idea of on-site meeting-room allocation at conferences, where attendees can sign up for 30-minute slots with speakers in an assigned room, avoiding travel to offices.
Anecdotal; the on-site meeting idea is an in-conversation suggestion.
No money is moving out of the bank account anywhere in the demo, which is why the low-accountability plan is still acceptable there.
The workshop's stated goal is building a financial analyst harness that can be trusted, so users stop 'talking with spreadsheets'.
Work with the model so it explains how your problem fits into its training dataset, reaching shared understanding.
described with lower confidence
When dictating to your phone and saying something in the middle of solving a problem, you can't take away that one sentence anymore and the whole analysis is gone.
Anekha Sokhal recommends that when a new technology wave arrives, you move very quickly to establish yourself in the field — her own approach was to launch multiple visible artifacts at once: the AI engineering post, a conference, and media.
Anecdotal strategy from the speaker's own experience.
Think about the trajectories of a user making a decision and look at what tools and information they use; verify what would have been the optimal choices or paths when aligning agent behavior with goals and examples.
endorsed with lower confidence
Felipe announces that Monzo has an in-house remote agent capability called 'agent chip', which will be covered in a breakout session the next day.
Anekha recounts being one of the very first at Ballyasny to use the ex-Firehose/Twitter Firehose data, describing it as a $300,000 per month data expense for hedge funds. The Firehose was free for researchers but hedge funds had to pay for it.
There is an element of the human experience where trust, accountability, and working together as a team matter a lot. Everyone has that friend who speaks very confidently about something they don't know — a counter-example to confident decision-making without evidence.
Anekha Sokhal's lesson from her own path: choose a career by matching your intrinsic interests — for her, a love of data and a fast pace — to the actual day-to-day work rather than prestige. After reading about trading, she 'could not imagine myself being in any other career.'
A generalization drawn from her personal experience.
Anekha Sokhal and the interviewee want to make the conference a high-signal conference and try to be the definitive thing next year in AI and finance.
medium
Anekha Sokhal distinguishes between people who have the problem but don't want to buy, and people who have the problem and want to buy — the matcher's job is finding the latter.
A generalization offered without supporting evidence in the atoms.
Moshi gets continuous hiring inbound from LinkedIn; one job posting had to be closed after four days with 2,500 applicants. The host endorses that $2,500 is too much and most of it is noise, so a better filter is needed. Many job applicants are legally not allowed to work (visa issues), adding noise to the applicant pool.
No published protocol matches this selection.
The creators behind most of what cleared.
344 cleared insights match
71 cleared insights match
Choose sources → to get a library of your own and a daily brief.
