Pinned Tweet
This pinned tweet will be used to save interesting notes on market microstructure as they are rare to come by, and we can learn a lot of first principles
By understanding the basics, we know what is considered normal, when anomalies arise, and therefore identify opportunities
By understanding the basics, we know what is considered normal, when anomalies arise, and therefore identify opportunities
1
0
11
1,334
Is there a reason why these models have free output? Are they copying Jev or there’s some optimal pricing going on
We're introducing the Gemini Decisions API in @GoogleAIStudio, powered by our new Gemini Decision 1 model.
Instead of generating a text response token by token, Gemini Decision 1 evaluates your input (across text, images, audio, video, and PDFs) against up to 1,000 questions to cisions in a single pass, with zero output tokens billed.
Here's a look at how it works 🧵
0
0
0
18
I’m learning a bit about what former verification is and I kind of understand why some of the nerds are saying that spec is more important.
Ironically, I had a brief stint as a product manager, and I wrote my product spec as this formal spec. No wonder why the engineers are so angry.
I didn’t get the memo ah haha
Ironically, I had a brief stint as a product manager, and I wrote my product spec as this formal spec. No wonder why the engineers are so angry.
I didn’t get the memo ah haha
1
0
0
150
When I was interviewing the outsourced interviewer told me to on my video but he didn’t turn it on himself
Then proceeded to ask me a lot of sensitive questions which I thought the application would have had it, istg if this is how I get a simswap attack this it it
Then proceeded to ask me a lot of sensitive questions which I thought the application would have had it, istg if this is how I get a simswap attack this it it
1
0
5
341
Genuinely cannot see myself going to web2. This place is mind poison but secretly you love it, do you really want to go free
0
0
2
53
Got cursor to audit some math in the morpho data warehouse by comparing against the morpho source code and tbh idk what its saying either
1) The tier 3 one was something I tried telling the AI was wrong for a long time and it kept calculating it wrong lmao.
2) No idea about e
1) The tier 3 one was something I tried telling the AI was wrong for a long time and it kept calculating it wrong lmao.
2) No idea about e
1
0
3
296
Man y’all WhatsApp sticker game is so weak how did you not discover copy and pasting a gif gives you a sticker tsk
0
0
0
144
I lowkey bag fumbled a random SQL question asked by an HR when asked:
"You have 2 tables, customer and orders table; find customers that didn't place an order."
I gave him the answer via pseudocode verbally. Kinda scared if I heard the “didn’t" wrongly; I was on full adrenaline mode.
"You have 2 tables, customer and orders table; find customers that didn't place an order."
I gave him the answer via pseudocode verbally. Kinda scared if I heard the “didn’t" wrongly; I was on full adrenaline mode.
3
0
5
767
I knew there could be a "technical" round, but not this early! Caught me off guard a little. Was practicing a lot more on my resume stories.
1
0
1
136
why is the code made by gemini so horrendous lmao, that's the rough shape of my code. but its actually not very good
0
0
1
124
Data Wolf 🐺 retweeted
"No matter your age, you'll always wish you started younger. But today is the youngest you'll ever start"
116
18,359
157,049
1,525,448
There are three grades of data in this business:
1) Raw data
2) Sector-normalized data
3) Protocol-specific data
Type 2 is actually really hard to make. Large data providers have no moat on type 3 datasets as an external party, since it can be internalized by the company itself (unless a VC needs it).
Over time it will likely become cost-prohibitive for the data company to maintain type 3, so they’ll keep it generic at type 2. This edge can be sustained if the firm adds coverage fast enough, but I’m unsure in the AI era.
The real competitor between type 2 and 3 is the analyst having to roll the grain up or down themselves because you don’t provide it or lacking the right dims_
The battle of the dataset lies in the last mile. They use case of the data will inform the shape of the data. Which is why it is kind of sad that the researchers are all fired because they are the best people to know and inform what the tables data analyst have to make, which then in turns help hedge funds ideate some new datasets to use. It creates a self fulfilling demand cycle.
1) Raw data
2) Sector-normalized data
3) Protocol-specific data
Type 2 is actually really hard to make. Large data providers have no moat on type 3 datasets as an external party, since it can be internalized by the company itself (unless a VC needs it).
Over time it will likely become cost-prohibitive for the data company to maintain type 3, so they’ll keep it generic at type 2. This edge can be sustained if the firm adds coverage fast enough, but I’m unsure in the AI era.
The real competitor between type 2 and 3 is the analyst having to roll the grain up or down themselves because you don’t provide it or lacking the right dims_
The battle of the dataset lies in the last mile. They use case of the data will inform the shape of the data. Which is why it is kind of sad that the researchers are all fired because they are the best people to know and inform what the tables data analyst have to make, which then in turns help hedge funds ideate some new datasets to use. It creates a self fulfilling demand cycle.
I find it exceptionally hard to believe there is a long-term business in providing publicly available blockchain data or derivations of that data to paying customers at any meaningful scale. Particularly when AI helps solve a bunch of the "messiness" that was alpha.
This is basically the hill Messari died on and to a lesser extent multiple other crypto research-type houses.
Its not sexy now but imho research, insights, and consulting for those moving into the space or looking to utilize the tech, are where you differentiate in this space long term. Huge open door for that with the recent business pivots and shutdowns.
Whos building it?
1
1
23
1,126
I remember during the flipside era, snowflake allows you to run AI models (zero shot style) and that was like the first innings of a jev like usage. But I wasn't that big on it because you can't control and tune the outputs in a classical data science style
0
0
4
214
Could remember the day he announced to team that he left for binance and I felt 🙁. He was like a ray of sunshine in a room
0
0
1
165
Genuine question, when you have a certain dataset or "insight", what does a chain actually do given that the flow is chain -> protocols -> users
As a chain how do you influence user behavior when its gated by the UI/UX/business of protocols and dapps?
Like what insight, say Solana has, then can affect pump fun's decision to do certain things to the user base (or this is completely out of scope?)
I mean I get solana level changes, because that's within the control of solana, but afaik it takes time to pass proposals, convince tokenholders etc.
Just curious to see some cool case studies
As a chain how do you influence user behavior when its gated by the UI/UX/business of protocols and dapps?
Like what insight, say Solana has, then can affect pump fun's decision to do certain things to the user base (or this is completely out of scope?)
I mean I get solana level changes, because that's within the control of solana, but afaik it takes time to pass proposals, convince tokenholders etc.
Just curious to see some cool case studies
surprised by how few teams do analytics to track their userbase
solana might be the only one that has a dedicated data team that takes it seriously. they analyse our top takers, makers, tvl holders
and more importantly how their user behaviour has changed in time
whoever focusses on this will win imo
2
0
7
521