51–60 of 518 posts
From the same source as the author cites, that's because their test runs are typically 5 miles and resuming manual control at the end of a test counts as a disengagement.
Didn't this just happen? Maybe my timescales are off, but I've been thinking about AI and Go since the late 90s, and plenty of real work was happening before then.
Outside a handful of specialists, I'd expect another 8-10 years before the current state of the art is generally understood, much less effectively applied elsewhere.
Earlier quoted context omitted.
The scaling argument in the article doesn't make any sense. There are rhetorical queries like "does this model with 1000x as many parameters work 1000x as well?" but what it means to scale or perform are not clearly or consistently defined - let alone defined in a way that would make your point about the utility of the advances. OpenAI's graph shows new architectures being used with more parameters because people are…
CFD was merely used as an example of something that does scale well. I'm not sure it was the best example, since CFD isn't very common. But basically you have a volume mesh and each cell iterates on the Navier-Stokes equation. So if you have N processor cores, you break the mesh in N pieces, each of which get processed in parallel. Doubling the number of cores allows you process double the amount in the same time, mi…
> Deepmind hasn't shown anything breathtaking since their Alpha Go zero. Didn't this just happen? Maybe my timescales are off, but I've been thinking about AI and Go since the late 90s, and plenty of real work was happening before then. Outside a handful of specialists, I'd expect another 8-10 years before the current state of the art is generally understood, much less effectively applied elsewhere.
Author's reasons: 1.Hype dies down (which is really good! Meaning the chance of burst, is actually lower!) 2.Doesn't scale is false claim. DL methods have scaled MUCH better than any other ML algorithms in recent history (scale SVM is no small task). Scaling for DL methods are much either as comparing to other traditional ML algorithms, where it can be naturally distributed and aggregated. 3. Partially true. But self…
Author here: I'm using deep learning daily so I have a bit of an idea on what I'm talking about. 1) Not my point. Hype is doing very well. But narrative begins to crack, actually indicative of a burst... 2) DL does not scale very well. It does scale better than other ML algorithm because those did not scale at all. If you want to know what scales very well, look at CFD (computational fluid dynamics). DL in nowhere ne…
A different take by Google’s cofounder, Sergey Brin, in his most recent Founders’ Letter to investors: “The new spring in artificial intelligence is the most significant development in computing in my lifetime.” He listed many examples below the quote. “understand images in Google Photos; enable Waymo cars to recognize and distinguish objects safely; significantly improve sound and camera quality in our hardware; und…
When I saw the Google demo of a CNN using video to split a single audio stream of two guys talking over each other, I became a believer.
Anyway i also don't get what the issue is with the model from radiology. It is already that good?! This is impressive. One model is close to well trained experts.
Just today i had an small idea for a new product based on what google was showing with the capabilities to distinguis two people talking in parallel.
At the last Google IO i was impressed because in comparision to the previous years, ML created better and more impressive products.
I was listing for years at key nodes about big data and was never impressed. I hear now about ML and im getting impressed more and more.
Earlier quoted context omitted.
The scaling argument in the article doesn't make any sense. There are rhetorical queries like "does this model with 1000x as many parameters work 1000x as well?" but what it means to scale or perform are not clearly or consistently defined - let alone defined in a way that would make your point about the utility of the advances. OpenAI's graph shows new architectures being used with more parameters because people are…
CFD was merely used as an example of something that does scale well. I'm not sure it was the best example, since CFD isn't very common. But basically you have a volume mesh and each cell iterates on the Navier-Stokes equation. So if you have N processor cores, you break the mesh in N pieces, each of which get processed in parallel. Doubling the number of cores allows you process double the amount in the same time, mi…
Earlier quoted context omitted.
The scaling argument in the article doesn't make any sense. There are rhetorical queries like "does this model with 1000x as many parameters work 1000x as well?" but what it means to scale or perform are not clearly or consistently defined - let alone defined in a way that would make your point about the utility of the advances. OpenAI's graph shows new architectures being used with more parameters because people are…
CFD was merely used as an example of something that does scale well. I'm not sure it was the best example, since CFD isn't very common. But basically you have a volume mesh and each cell iterates on the Navier-Stokes equation. So if you have N processor cores, you break the mesh in N pieces, each of which get processed in parallel. Doubling the number of cores allows you process double the amount in the same time, mi…
- Improvement is hard to define consistently. Sometimes, improving classification accuracy by 0.5% means reducing error by 20%, and makes economic applications that have 100x the value or frequency of use.
- Resources used in training can be amortized over billions of times the same model is reused (much more cheaply). So even achieving an epsilon improvement in the expected utility of each inference can justify a massive increase in training cost.
- Some other notions of "better results" or "less expensive" include amount of training data required, social fairness of results, memory required or power used during inference, and so on. And there are major advances in current research on each of these better formalized axes!
That last bit is what is so frustrating in reading an article like this. The author is sweeping aside with vague arguments a great deal of work that has been written and justified to a much much higher standard of rigor (not just the VCs we all like to snark about). Readers should beware of trusting a summary like this without engaging directly with the source material.