feels like it’s an annual tradition at @delian_ai to be installing systems for customers in remote locations during the festive season. when i first met the team, they’d just been doing the same! security never sleeps.
SYSTEMS
-
Designing Machine Learning Systems for Production-Ready Applications
By
–
Designing #MachineLearning Systems — An Iterative Process for Production-Ready Applications: http://
amzn.to/46epLSi by @chipro (formerly @ClaypotAI
, now @VoltronData
)
———
#DataScience #AI #ML #MLOps #DataScientist
———
Topics covered: Engineering data and choosing the right -
Intentional Design Complexity: Modularity and Hierarchy Limits
By
–
Agreed on the limits of biological evolution, but why do intentional processes hit a complexity limit? Through modularity, hierarchy, and recursion, there doesn’t seem to a limit on intentional design.
-
AI Models Improving at Explaining Reasoning for Human-Machine Collaboration
By
–
AI models are getting better at explaining their reasoning.
— Erik Brynjolfsson (@erikbryn) 20 décembre 2024
This will have lots of benefits, especially when cooperating with humans to create human+machine systems that are better than either human or machine separately. https://t.co/WY9sXSJPMkAI models are getting better at explaining their reasoning. This will have lots of benefits, especially when cooperating with humans to create human+machine systems that are better than either human or machine separately.
-
Autonomous drones and robots eliminating distance constraints
By
–
Quite amazing! The next step is fully autonomous drones and robots, so the distance won't even matter.
-
Managing Uncertainty: Balancing Reflexive and Reflective Responses
By
–
16/25 How do we manage the uncertainty? I think the answer is through balancing reflexive and reflective responses. Here's today's thoughts and Dan Andrew's capture of them for Chapter 7. https://
the-atomic-human.ai/reflections/ba
lancing-reflective-and-reflexive/
… -
10 Keys to Unlock GenAI in the Enterprise
By
–
10 Keys to Unlock #GenAI in the Enterprise Prompt engineering System integration Performance Scalability Response validation Context management Bias & ethics User personalization Data privacy Continuous learning #AI
-
All Models BF16: Context Length Maximization via TGI v3
By
–
All models are bf16. re: context, I know in past we have had restricted usage during spikes, would need to double check for 3.3. We do try to maximise the context length where possible. btw all powered via TGI v3:
-
SCHED_FIFO Linux Scheduling vs CUDA Stream Priorities
By
–
I love SCHED_FIFO on Linux not being wish-washy about scheduling — if a higher priority can run, it preempts anything lower Right Now, and it won’t stop until it yields, or something even higher takes it. CUDA stream priorities are more like traditional OS priority hints. Better
-
Integrated Large Models Combined with Good Old Fashioned Engineering
By
–
New computer systems perspective on integrated (modular and hybrid) large models with good old fashioned engineering (GOFE). @UCBerkeley @Berkeley_EECS @Databricks https://
arxiv.org/abs/2412.05299