Medha Basu
@medha_basu
Followers
1K
Following
2K
Media
175
Statuses
2K
Co-founder @DefogData (YC W23). Former journalist/editor/content marketing person. Once interviewed a terrorist
Singapore / San Francisco
Joined January 2013
Asked Claude Code to use an LLM to summarize stuff for a CLI tool I'm building. Why did it decide to use OpenAI 😆
3
1
7
Using Claude Code is insane. I feel very powerful. Like I have superpowers I really should not
0
0
3
# the nightmare bicycle imo, the most important idea in product design is to avoid the "nightmare bicycle". imagine a bicycle where the product manager said "people don't get math so we can't have numbered gears - we need to have labeled buttons for gravel mode, downhill mode,
120
292
3K
Open-sourcing Introspect: MIT-licensed Deep-Research for your internal data! Works with spreadsheets, databases, PDFs, and web search. Has a remarkably simple architecture – Sonnet agent armed with recursive tool calling and 3 default tools. Best for use-cases where you want to
18
100
660
"You can run it on a single L4 GPU that costs $300/mo on GCP" "You can use logprobs and attention scores to determine where, exactly is the model paying attention to inside a prompt + what it's getting confused by when generating outputs." Very cool! congratulations on the
We made a thing! Very happy to announce sqlcoder-pro and the Defog Alignment Platform. Available to use immediately without a wait-list, weights will be open-sourced very soon. The video does a quick show and tell comparison against ChatGPT (with gpt-4o). Read on for more
0
1
3
We made a thing! Very happy to announce sqlcoder-pro and the Defog Alignment Platform. Available to use immediately without a wait-list, weights will be open-sourced very soon. The video does a quick show and tell comparison against ChatGPT (with gpt-4o). Read on for more
10
23
123
Llama-3 based SQLCoder 8b is out! Open weights with a commercially friendly cc-by-sa license. Probably the best <10B param model for Postgres text to SQL right now. Slightly better than gpt-4-turbo and claude opus for 0-shot text to SQL generation. Also approaches their
27
117
580
7B model better than the most recent GPT4 out of the gate on text to SQL! I've said it and will say it again: except for very general tasks (like a search engine), smaller models are better/cheaper/faster!
Launching the second generation of SQLCoder-7b on @huggingface today! This is distilled from our 70B model, and performs around as well* as GPT-4 for text-to-SQL generation. Finetuned on @AIatMeta's CodeLlama-7b. *To be more precise – this model is much better at ratios and
6
18
167
it works! SQLCoder70B for MLX / apple silicon is live. the model path is here: https://t.co/9ukO5EaEEy you'll need defog's system prompt below for it to work. (question and response outputs a bunch of junk tokens but outputs work well for new / fresh convos): ### Task
4
8
42
And incredibly grateful to @JP_smasher and @wendyaww without whom this would never have happened. Incredibly lucky be working with these rockstars!
2
3
12
We just opened sourced SQLCoder-70B! It outperforms all publicly accessible LLMs for Postgres text-to-SQL generation by a very wide margin. SQLCoder is finetuned on @AIatMeta's CodeLlama-70B model that was released yesterday on less than 20,000 hand-curated prompt completion
58
269
2K
Welp, just finished training and evaluating CodeLlama-70B for SQL. This thing is a beast when fine-tuned. Miles ahead of anything else (including GPT-4). Open-sourcing the weights either today or tomorrow!
35
110
2K
Nice to see Defog on the lists for top 100 YC startups in Generative AI, Enterprise and Open Source!
2
6
31
Way to go @defogdata! ⚡️ So so bullish on what @medha_basu & @rishdotblog are up to.. Open source is the way! 🤗
In the past month, we caught up with fast-growing @stripe users like @yongfook, @medha_basu, @kiwicopple, and @ytk141 to hear how 2023 went for them. Excited to finally share this with the world today. ✨
0
2
10
Running our new 7B model 100% locally on an M1 Mac 🤓 76% accuracy on sql-eval with GGUF. For reference, GPT-4 is 82.5% and SQLCoder-34B-v2 is at 85% Pretty wild that this works locally *on a laptop*! Super excited about getting this on a Mac app soon.
28
61
855
@rishdotblog @ajhodls @ycombinator Thrilled to be on board as a #happyinvestor! Look forward to watching you build an amazing company. Cheers.
2
2
22
The @defogdata team moves at light speed. We’re thrilled to partner with them to bring the most powerful open-source LLMs for enterprise data analysis. @rishdotblog and @medha_basu are exceptional founders and headed for big things. 🚀
We are excited to announce Defog’s $2.2M funding round to develop open-source LLMs for data analysis. https://t.co/V360ID50rD The round was led by Script Capital (@ajhodls) and @ycombinator, with participation from Hike Ventures, Pioneer Fund, and notable angels. Since launch,
2
3
20
We are excited to announce Defog’s $2.2M funding round to develop open-source LLMs for data analysis. https://t.co/V360ID50rD The round was led by Script Capital (@ajhodls) and @ycombinator, with participation from Hike Ventures, Pioneer Fund, and notable angels. Since launch,
4
6
48
I will say this: I have personally tested all models listed here and many more, including ones that claim to reach GPT4-level results for natural language to SQL over generic benchmarks. But the Defog model is the only one I've seen thus far that actually delivers.
1
1
9
Can't wait to share more on the partnership at some point, but for now I will say that this is a result of the incredible work by my amazing colleagues this year ❤️
0
0
9