HNHacker News
TopNewBestAskShowJobs

austin-starks

31 karma · joined August 24, 2023

submissionscomments
austin-starks··on [dead]
It's not "the future". It is happening RIGHT NOW.

Cope... or use it for yourself

austin-starks··on [dead]
TL;DR, for ambiguous tasks, Claude 3.5 Sonnet performed far better than the other language models. It generated a SQL query that better conformed to the user's input, and created a JSON object that demonstrated superior reasoning skills. These tasks were related to highly-specific finance tasks.
austin-starks··on [dead]
Key Points:

In a task of generating SQL queries, I found that Google Gemini Flash 2.0 was much better than DeepSeek R1 and OpenAI o3-mini. Specifically, I found that it was very significantly cheaper, had a much higher context window, lightning fast, and was more accurate when it came to generating syntactically-valid SQL queries

austin-starks··on [dead]
Key Points:

In a task of generating SQL queries, I found that Google Gemini Flash 2.0 was much better than DeepSeek R1 and OpenAI o3-mini. Specifically, I found that it was: * Very significantly cheaper * Had a much higher context window * Lightning fast * Was more accurate when it came to generating syntactically-valid SQL queries

austin-starks··on I used OpenAI's O3-mini to create a trading strategy. It's DESTROYING the market
I wrote this step-by-step guide on how to use O3-mini to create, test, and deploy an algorithmic trading strategy. In my test, I showed that the strategy created by O3 beats the market in every metric, including sharpe ratio, sortino ratio, and percent change.

What are your thoughts?

austin-starks··on [dead]
TL;DR: OpenAI's o3-mini model is faster, more reliable, and more accurate than DeepSeek's R1
austin-starks··on I am among the first to gain access to OpenAI's "Operator". Here are my thoughts
TL;DR: OpenAI's Operator, an AI-powered agent designed to autonomously interact with web browsers to accomplish tasks, fell short of expectations during testing. While it initially impressed with its ability to navigate and gather information, its limitations became apparent through hallucinations, lack of efficiency, and failure to request user assistance when needed. Despite its potential, Operator is not yet ready for real-world productivity, but future improvements could address these issues.
austin-starks··on China won A side-by-side comparison of DeepSeek R1 vs. OpenAI O1 for Finance
It matters because of the ridiculous regulations imposed by the US on China. Even despite them, they've built a better model.

I HIGHLY doubt it was built for these type of operations.

austin-starks··on China won A side-by-side comparison of DeepSeek R1 vs. OpenAI O1 for Finance
I couldn't agree more. China has: * weaker GPUs * a smaller model * started with nothing

And now they're building better, faster, and cheaper models at a fraction of the cost. It's hilarious and exciting.

austin-starks··on China won A side-by-side comparison of DeepSeek R1 vs. OpenAI O1 for Finance
Facts. OpenAI thought they had a moat, but even with the shittiest of GPUs, China managed to developer a better, faster, and cheaper mode, AND they open-sourced it.

With that being said, Llama 4 will be CRAZY

austin-starks··on I created an app that helps investors deploy no-code algotrading strategies
I created an app that teaches retail investors how to become Wall Street grade traders. Specifically, I have step-by-step tutorials on how to do the following:

* Create an algorithmic trading strategies * Test them on historical data (a process called backtesting) * Testing them in real-time risk-free

I've already received lots of positive feedback and wanted to share it with the community and ask for more. If there's any tutorial that you think would valuable, please let me know!

← PreviousPage 2 of 2