Ben Hamner
@benhamner
Followers
32K
Following
19K
Media
735
Statuses
5K
working on something new. formerly @kaggle cto
SF
Joined March 2009
Wordle xxx 1/6. 🟩🟩🟩🟩🟩. Get Wordle right on your first guess using the daily ⬛🟨🟩 tweet distribution
33
221
1K
Buzz: AI is going to imminently replace software engineers. Reality: all the latest models invent duckdb json functions when trying to extract fields from json data in a duckdb file.
2
0
2
Getting a bunch of 429 rate limit errors with a quota exceeded message from @OpenAI's API, in spite of being well under quota and published rate limits.
2
0
5
NLC: natural language cron.
We're excited to introduce Tasks! For the first time, ChatGPT can manage tasks asynchronously on your behalf—whether it's a one-time request or an ongoing routine. Here are my favorite use cases:. 1/ ChatGPT checks stock price every morning!
0
0
2
Just had the most obnoxious interaction I’ve had in a long time. Jogging in quiet Palo Alto neighborhood. Middle-aged man and woman block away walking towards me. Half a block away they start screaming “RUN ON THE FUCKING STREET SIDEWALKS ARE FOR WALKING NOT RUNNING. GET OFF THE.
8
0
15
Marimo's a delight to use for both exploratory Python notebooks and rapidly prototpying reactive data applications with a rich UI. It's replaced Jupyter notebooks for me. Excited to see what Akshay and team build from here!.
My co-founder @themylesfiles and I have started Marimo Inc. to keep building the @marimo_io notebook and other Python data tools. We've raised a $5M seed round led by @antgoldbloom and @shyammani_ at @aixventureshq. Excited for the journey ahead!.
1
0
12
RT @antgoldbloom: Looking at job posts over the last year, it looks like Amazon is making the broadest AI investment. 228 teams across the….
0
15
0
Fast hack to improve the factual accuracy of your LLM calls? Multi-LLM and consensus (e.g. majority voting) across them among the leading foundation models.
2
0
5
RT @antgoldbloom: The overlooked GenAI use case: cleaning, processing, and analyzing data. Job post data tell us….
0
119
0
10 data quality challenges in relational tables:. 1. Duplicates.2. 3. Duplicates.4. Cḧarαctěr Énçødïng.5. 1/1/24 (ambiguous dates).6. 7am (ambiguous times w/o timezones).7. John A. Smith, john smith (fuzzy joining).8. asdfasdfa junk record.9. Company: Apple Url:.
0
0
8
Close to 100% of machine code is now compiler-written. Higher levels of abstraction are how we move forward, faster.
0
0
8
I was born into the first computer-native generation. My 2yo son’s in the first AI-native generation.
0
0
2
Admin consoles have their own circle of UX hell. If you can read this Google GSuite admin setting and tell me the difference between "Trusted" and "Limited", your English understanding is much better than mine
3
0
6
For all the marketers that want to slap AI on a product name: as soon as I see "AI" in a product, I assume it's not going to work as advertised.
0
1
7
This. Breaking through a lot of default bureaucratic process where appropriate critical to integrating Kaggle effectively into Google.
I've consistently seen that organizations run by process fail, while organizations that win are run by a handful of opinionated individuals.
0
1
7
Fine-tuning with the right data for your use trumps selecting exactly the right model. And can make a tiny model outperform a larger generalist model with lower latency and cost. @kaggle datasets covers a huge range of niche use cases as starting points for finetuning.
Kaggle datasets ( are one of the most under-discovered resources in machine learning. There are 367K datasets covering a wide range of niche domains. A micro-trend I'm seeing is people using these datasets for fine-tuning (or another round of.
0
6
16
100% something that’s not going to do what they say it’s going to do.
If someone says they’ve built an “agent”, do you know what they mean by that?.
2
1
14
Working in ML right now is like having Christmas or a birthday every day
Today, we release Mistral Large 2, the new version of our largest model. Mistral Large 2 is a 123B-parameter model with a 128k context window. On many benchmarks (notably in code generation and math), it is superior or on par with Llama 3.1 405B. Like Mistral NeMo, it was trained.
0
1
24
If you know talented engineers and data people that you think would be a great fit for us, would appreciate referrals / sharing these links with them.
4
0
3