@ryanmhickman Would love to test one of those for latency. I've recently been comparing a lot of cameras for their inherent, camera-to-computer latency and found that that's where most of the the glass-to-glass remote latency comes from (e.g., 100 ms for USB). Now I'm hunting for faster ones!
@samanfarid It's not a flywheel! I know what you mean, but the flywheel analogy is wrong -- even if everyone is using it. Snowball is a better analogy: it picks up more and more snow as it grows and rolls faster.
@karpathy Do you think that input/output mind-meld would perhaps be better supported by something like Hinton's deep believe networks rather than feed-forward NNs? The argument being that melting input and output will be easier when using the "same vocabulary".
@simonkalouche The only reason this seems impressive is because people anthropomorphize humanoids (for obvious reasons) -- they imagine a human doing this. From a control perspective this is barely more impressive than an inverted pendulum.
@adrianmacneil@AlexRoy144@Tweetermeyer@kirstenkorosec That "data flywheel" metaphor needs to die! It's simply wrong. It doesn't represent what is intended. A flywheel is an energy _store_. You never get out more than you put in. Which is the opposite of what people are trying to say with the flywheel. They mean an upward spiral.
@BradPorter_@bznotes So it's not the smartest of founders that survive, but the ones most adaptive to change? -- Another good reason to stay small and nimble for as long as possible!
Transitive 2.0 is here! Adding MQTT-history storage in ClickHouse, visualization in Grafana, and alerting via Alertmanager. See these new features in action in the updated Health Monitoring capability.
https://t.co/BxmwdmjNKj
#clickhouse#robotics#grafana
@adrianmacneil > hand-crafted code is like manually writing assembly
That analogy doesn't quite work, because higher-level languages like C compile _deterministically_ to assembly/machine-code. AI coding still lacks this determinism.
AI is biting the hand that feeds it. Most obvious evidence: the drop in questions on StackOverflow after ChatGPT came out. How will AI get better if it replaces its own source of training data?