Community of practitioners working to promote and encourage the adoption of successful ML projects. Organizers of TMLS in June & @MLOpsWorld Nov, Austin TX.
Toronto, Ontario
Joined August 2017
- Tweets2.2K
- Following2.2K
- Followers4.8K
- Likes2.7K
From “generate text” to “make a decision”
The result is speed and cost gains of an order of magnitude (not a small percent).
→ Recursive Self-Improvement track. MLOps North, Thursday November 5, 3:55
torontomachinelearning.com/
Toronto Machine Learning Society (TMLS) retweeted
Over the last few months, and especially the last few weeks, we've seen a lot of news about AI running amok. There's been a lot of hype, and some misguided opinions. But also some grounded advice. and I have been discussing how @TMLS_TO a voice in the cacophony.
It is possible for Generative AI at scale to do things it was not designed to do, things it was never intended to.
Part 2 of Lies, Damned Lies and LLMs maps eight risks and five opportunities onto agentic systems.
tmlsinsights.substack.com/p/…
Deception is a human concept. It requires motivation and intent.
Can an LLM deceive? “Philosophically... maybe? Practically - probably.”
Lies, Damned Lies and LLMs, Part 1, by Graham Toppin and David Scharbach.
tmlsinsights.substack.com/p/…
When a model shows you its reasoning, what you see is a summary of the model’s internal trace, not the raw reasoning.
New in TMLS Insights: chains of thought, reasoning traces and working memory, and what they mean for your pipeline.
tmlsinsights.substack.com/p/…
Radium had to decide which layers of the AI stack it needed to own, and which it could responsibly source globally.
Adam Hendin (Radium) says the decision shortened the company’s path to market.
AI Sovereignty track. MLOps North, November 5, 1:30 p.m.
torontomachinelearning.com/
What happens if you stop treating coding agents as unreliable code generators?
Geoffrey Hunter (Curinos) found development velocity improves while quality holds.
Coding Agents and Dev Workflows track. MLOps North, November 5, 2:55 p.m.
torontomachinelearning.com/
What happens if you stop treating coding agents as unreliable code generators?
Geoffrey Hunter (Curinos) found development velocity improves while quality holds.
Coding Agents and Dev Workflows track. MLOps North, November 5, 2:55 p.m.
torontomachinelearning.com/
Teams wanting to deploy on their own hardware usually face a wall of Kubernetes complexity.
Alex Tucker (49ops) built Lairkit to establish a self-hosted deployment platform on any hardware.
AI Sovereignty track. MLOps North, November 6, 1:30 p.m.
torontomachinelearning.com/
The usual assumption is that giving an agent more autonomy means giving up control.
Millin Gabani (Worklayer) argues that well-designed feedback and explicit limits do the opposite.
Recursive Self-Improvement track. MLOps North, November 5, 3:55 p.m.
torontomachinelearning.com/
Kiana Haddadi is a Data Scientist at BMO.
When her team built a skill to generate model documentation, the obvious design was a productivity tool.
Instead, they designed the skill to treat information as claims to be proven, rather than ground truth.
torontomachinelearning.com/
Code is now produced faster than it can be reviewed.
Aditya Shukla (AI Engineer, Georgian Partners) found a creative way to test it against the specification rather than the implementation.
He’s presenting this at MLOps North, Toronto. lnkd.in/g6EEsEc
Can we reframe “Loop Engineering” as applied cybernetics?
Sayantan Das (Senior Applied AI Scientist, Manulife) is making the case that a decade of Kubernetes production learnings transfers directly.
→ MLOps North, Thursday November 5
torontomachinelearning.com
Out of 100+ enterprise AI leaders on our committee, evals were the single biggest blocker for putting agents in production.
→ MLOps North, Friday November 6, 11:45 a.m., RBC WaterPark Place, Toronto.
torontomachinelearning.com
Wisedocs automated large portions of model development across 1500+ document types.Aryan Dhar (Senior ML Engineer, Wisedocs) and Burak Yildiz, Ph.D. are showing what they built.
→ MLOps North, Thursday November 5, RBC WaterPark Place, Toronto.
torontomachinelearning.com
An answer from an LLM can be right and still be unusable.
This can happen if the evidence it provides is stale, contradictory, or taken from the wrong source.
→ MLOps North, Thursday November 5, 10:55 a.m., RBC WaterPark Place, Toronto
torontomachinelearning.com
