Divyanshu Abhichandani

Software engineer @Harvey

[Gabe Pereyra @gabepereyra]

Update on @harvey’s model training effort.

We post-trained a model we are calling Tenet:
- Achieves SOTA on LAB
- Generalizes 3rd party legal benchmarks
- Uses sub-agents for domain specific capabilities

Tenet uses Kimi K3 as base and was post-trained in collaboration with @FireworksAI_HQ:
- Rank-64 LoRA over the full network
- GSPO with importance-ratio masking
- 134 B300 GPUs for 2 months

Despite not being trained on 3rd party legal datasets we found improvements on:
- @mercor’s Apex Agents - Corporate Law
- @crosbylegal’s Redline Bench
- LegalBench

Tenet also learned how to use domain-specific subagents (separate post-trained models) for complex tasks:
- M&A Diligence: training in an RLM harness for long-horizon tasks (with @baseten)
- Review Table: specialist models for high-volume structured data extraction (with @appliedcompute)
- Firm Knowledge: parametric memory and structured notes for more efficient enterprise search (with @engram)

These results suggest we can significantly scale training and we plan to:
- Scale both human and synthetic data significantly and scale training to 1K and then 10K GPUs
- This scale will let us move to full parameter fine-tuning and larger models
- We are now starting to post train models in our production harnesses
- Post-train other open-source base models to provide customers with model choice

If these problems sound interesting we are hiring for our post-training team

[Winston Weinberg @winstonweinberg]

Proud to finally announce that Goldman and JPM are backing Harvey.

[Divyanshu Abhichandani @divyanshu07]

Proud to see Microsoft's legal team adopt Harvey.

[Harvey @harvey]

Microsoft’s Corporate, External, and Legal Affairs (CELA) organization will use Harvey across its legal and compliance operations.

[Julio Pereyra @ItsJulioPereyra]

Environments to train agents for diligence is one of my current favorites of our ongoing research projects.

A few reasons why:

[Divyanshu Abhichandani @divyanshu07]

These days, ‘learning a new skill’ just means editing skill.md 😅

[Winston Weinberg @winstonweinberg]

With @FireworksAI_HQ we’re researching agent routing to more intelligently navigate quality-cost trade-offs for long-horizon work

[jack @jack]

greatest bitcoin miner of all time: https://t.co/sZUp23k3VT

[Sanket Shah @_sankyy]

Here’s the raw breakdown of how-we created our product release video. Clear-cut answer to all your anxieties with AI video. I’m listing down what worked, the messy middle of it all.
Check the thread:

[Divyanshu Abhichandani @divyanshu07]

@_sankyy 🔥 🔥

[Cute Baby @nachunja]

$XYZ – New @Square handheld device

It's beautiful 👏👏👏

[jack @jack]

today is @square’s biggest release in our company’s history.

over 100 products and features. including the start of SquareAI.

we are collecting all of them, and will keep our roadmap updated, here: https://t.co/t0hkrgXfaJ

[Divyanshu Abhichandani @divyanshu07]

@TimesAlgebraIND @CommunityNotes

[jack @jack]

goose never died…and is #1 trending on github

https://t.co/Mss7abGodq

[jack @jack]

a new local open source agent, to help with all your eng tasks

[Divyanshu Abhichandani @divyanshu07]

@wholemars how long do I wait for v12? Still on v11.4.9

[Chris Bakke @ChrisJBakke]

San Francisco has the highest % of people who need 45 minutes of Calm meditation, 25 minutes in an Equinox sauna, a $17 smoothie, and 8.5 hours of rest in their Eight Sleep to recover from filing 4 Jira tickets but also want to leave this high-pressure life and become a farmer.

[Divyanshu Abhichandani @divyanshu07]

@DsMatie 👋🏻

[Divyanshu Abhichandani @divyanshu07]

@SakshiSuthar28 @CGISFO @DrSJaishankar @MEAIndia please help.

[Sanket Shah @_sankyy]

What You Type Is What You Get! @InVideoOfficial AI coming soon! Check it out!! 🤯🤯🤯

[Divyanshu Abhichandani @divyanshu07]

@sanket_invideo @InVideoOfficial 🔥🔥