M37Labs Logo

M37Labs Feed

Knowledge base

Direct from our engineering and research Slack channels. Raw insights, technical breakdowns, field notes, and real-time updates from the M37Labs team.

M

Team M37Labs

knowledge-base
8 months ago

TOON (Token-Optimized Object Notation):

https://github.com/toon-format/toon
TOON (Token-Optimized Object Notation):
it’s a data format I'm testing that makes AI much more efficient. Instead of repeating labels (like name or email) for every single item, it defines them once at the top, kind of like a spreadsheet. This means the AI has to read fewer characters and tokens.

I just finished benchmarking JSON against TOON across several Google Gemini models, and the result :
I found that TOON reduces token usage by 30–40% for lists, logs, and sensor data, This simple switch can cut our LLM API costs by nearly one-third for most data-heavy tasks.
image.png
Expand Image
ThreadImageCode
Mar 24, 2026 • 12:51 PM
M

Team M37Labs

knowledge-base
8 months ago

Mamba-3: Improved Sequence Modeling using State Space Principles

https://arxiv.org/abs/2603.15569
arXiv.org

Mamba-3: Improved Sequence Modeling using State Space Principles

Scaling inference-time compute has emerged as an important driver of LLM performance, making inference efficiency a central focus of model design alongside model quality. While the current Transformer-based models deliver strong model quality, their quadratic compute and linear memory make inference expensive. This has spurred the development of sub-quadratic models with reduced linear compute and constant memory requirements. However, many recent linear models trade off model quality and capability for algorithmic efficiency, failing on tasks such as state tracking. Moreover, their theoretically linear inference remains hardware-inefficient in practice. Guided by an inference-first perspective, we introduce three core methodological improvements inspired by the state space model (SSM) viewpoint of linear models. We combine: (1) a more expressive recurrence derived from SSM discretization, (2) a complex-valued state update rule that enables richer state tracking, and (3) a multi-input, multi-output (MIMO) formulation for better model performance without increasing decode latency. Together with architectural refinements, our Mamba-3 model achieves significant gains across retrieval, state-tracking, and downstream language modeling tasks. At the 1.5B scale, Mamba-3 improves average downstream accuracy by 0.6 percentage points compared to the next best model (Gated DeltaNet), with Mamba-3's MIMO variant further improving accuracy by another 1.2 points for a total 1.8 point gain. Across state-size experiments, Mamba-3 achieves comparable perplexity to Mamba-2 despite using half of its predecessor's state size. Our evaluations demonstrate Mamba-3's ability to advance the performance-efficiency Pareto frontier.

arxiv.org

Link
Mar 23, 2026 • 12:30 PM
M

Team M37Labs

knowledge-base
8 months ago

Venture capital is flowing into artificial intelligence-driven wealth management startups…

Venture capital is flowing into artificial intelligence-driven wealth management startups as middle-income Indians move beyond gold and real estate into financial assets.

New thesis: Startups are using AI to deliver lower-cost, personalised advisory, opening up services once limited to high-net-worth individuals.

For instance:

• Otto Money raised $1.3 million last month and is building an AI-led mutual fund advisory chatbot; it is now eyeing a $10 million raise.Gurugram-based daily savings app Bachatt is in talks to raise $12 million, likely led by Accel.Credit management startup Oolka is also looking to closea $12 million round.
The Economic Times

AI-led wealthtech rising as VCs bet on a fintech disruption - The Economic Times

As middle-income Indians increasingly diversify beyond real estate and gold into financial assets, the investment thesis is evolving. New-age startups are leveraging AI to build products at lower cost and with greater efficiency.

nltrack.indiatimes.com

Preview image for AI-led wealthtech rising as VCs bet on a fintech disruption - The Economic Times
The Economic Times

Accel set to lead $12 million round in AI fintech Oolka - The Economic Times

The round comes on the back of a $7 million infusion led by Lightspeed and Z47 last September. The company counts Livspace founder Ramakant Sharma, Meesho’s Vidit Aatrey, and Blackbuck’s Rajesh Yabaji as angel investors.

nltrack.indiatimes.com

Preview image for Accel set to lead $12 million round in AI fintech Oolka - The Economic Times
Link
Mar 23, 2026 • 10:39 AM