sumnuyungi

This is such an HN comment: pedantic, emotional, arrogant, completely out of touch with reality.

> I got a snotty response that I wouldn't be able to smell things that are in a sealed package, and I should always go by the manufacturing date printed on the package. No. You're wrong. And you're wrong about food safety, which is really not good. You go by smell, sliminess, color. The printed date is a nice guide but NOT what anyone should be using for food safety.

Did you consider that meats are sealed in store and you cannot smell it before purchasing? Going by date is fine for most areas. I also don't understand how you can be upset over "overly-patronizing" instructions but also want the level of detail for inspecting meat to include texture/smell/appearance.

> Now I've completed the food safety and "do you know what lean meat is" quiz, and I immediately go into another quiz about buying cheese. It wants me to say which cheese is hard and how to store cheese. What does this have to do with cooking a cheeseburger? I gave up.

Ingredient preparation is a basic component of any recipe. Knowing how to store cheese has a pretty obvious relationship to cooking a cheeseburger.

> It seems like a knowledge graph the way you've done it is an anti-pattern of actual cooking. You're decomposing every recipe into its ingredients and then asking arbitrary questions about each one.

It seems like if you think cheese storage is arbitrary, you do not cook for yourself on a regular basis enough to think about ingredient storage/preparation.

> I don't think a framework of focusing on ingredients in a vacuum, having people get quizzes wrong, and calling them "Junior Chef" is the way to do it.

Knowing the ingredients you're working with is the basic foundation to cooking. I don't see how someone can try to make a "Duolingo for Cooking" without including foundations... it'd be like trying to teach Spanish without defining pronouns.

Because analysts have always been so accurate forecasting Tesla...

We already have the witness list and know roughly what to expect. The main thing going against Amber is that we've heard from the officers that responded to a couple incidences and each officer has testified as to not observing any injuries nor any damage to the penthouse. There are also zero medical records reporting the injuries she claims to have sustained.

Her witness list does not include anyone that will refute those points and is mainly filled with people whose only account of what happened between Johnny/Amber is what they've heard from Amber.

Just a sample of her claims thus far:

> Walked across a tile floor covered in glass shards from broken bottles and wine

> Sexual assault with a (potentially broken) wine bottle

> Thrown across a room by the neck

> Beaten on top of a bed so forcefully as to have broken the timber bed frame

In Heard's case, the ACLU is creating an ambassador position and crafting the credentials for the ambassador they wish to hire (the op-ed). Knowing the level of deception they're engaging in - and not from a low-level employee but from their general counsel - how do you trust/verify any claims the ACLU makes?

You can impeach someone and/or offer up other witnesses to chip away at a liar's credibility. "Good enough" doesn't come close to describing the level of deception necessary to render our entire legal system completely ineffective.

The article's phrasing is a mischaracterization: Depp and Musk made donations on behalf of Heard, the ACLU didn't go after them to "collect".

> The Swiss data seems pretty conclusive in favor of my point, as it includes the omicron wave.

1. You can't use the entire range back to Jan 2021 to calculate current vaccine effectiveness and 2. if you use data from Nov 2021 (Omicron start), the Swiss data presents non-linearity that suggests small sample size/incomplete data.

> Also, see the recent nyc hospitalization data, which includes the entire omicron wave (now fading): https://coronavirus.health.ny.gov/covid-19-breakthrough-data. Note, this raw data understates vaccine efficacy as a relatively high proportion of vaccinated individuals are vaccinate with rather than for COVID.

It understates vaccine efficacy or overemphasizes omicron severity. At this point, we have enough data from regions with varying vaccination rates to conclude that it's the latter.

> Not sure why the age bracket point is relevant unless you think the vaccines have different relative efficacy within different age brackets for omicron - that was not the case for prior variants.

I selected age brackets for time ranges during Omicron wave and the spikes suggest small sample sizes. Small sample sizes can't be used to draw any conclusions from re: vaccine effectiveness.

It really is not a disputed claim that 2 shots do not provide much protection over the baseline for Omicron. Hence this post submission with Pfizer launching an Omicron-specific shot...

Omicron is the dominant variant so you have to throw out all data prior to Nov 2021. When I use different age groups in the link you provided, the numbers imply a pretty small sample size. No deaths since Nov among 39 or younger, strange nonlinearity for 40-59.

Not the study from Reuters that I linked, but another with a specific definition of "vaccine effectiveness":

During the proxy omicron period, we found a vaccine effectiveness of 70% (95% confidence interval [CI], 62 to 76), a finding that was supported by the results of all sensitivity tests. This measure of vaccine effectiveness was significantly different from that during the comparator period, when the rate was 93% (95% CI, 90 to 94) against hospitalization for Covid-19 (Table 2).

[1] https://www.nejm.org/doi/full/10.1056/NEJMc2119270#:~:text=D....

SSI paper had Pfizer/Moderna 2 shots at 55.2%/36.7% for Omicron and waning quickly enough to require a booster within 5 months of the last shot.

[1] https://www.reuters.com/business/healthcare-pharmaceuticals/...

Compared to what? Their pricing is in-line with the previous Intel models and are a huge upgrade in performance. For some tasks, you have to look at a desktop configuration for comparable performance (at a much higher price).

To finally use my laptop as it's intended: without being tethered to an outlet. The battery life on these is phenomenal.

I went to your personal website and would recommend updating the about page.

Do you have a non-paywall link to that article or a link to the actual data? The article should list sources.

The toxicity on HN is a strange combination of vitriol and superiority complex. I can't really think of many other forums that spew such hatred while morally grandstanding to such an extent.

What radar did Tesla use? What does the rest of the industry use?

Didn't Andrej show examples of emergency braking and how poorly radar performed v. vision?

You make it sound extremely manual and sequential when reality is anything but.

A team with funds like Tesla, Google, FAIR is going to be using NAS and have a continuous testing pipeline. Tesla has arguably the best environment for continuous testing which is the most difficult part of improving a model. Andrej even said in his talk that their supercomputer is in the top 5 for FLOPs.

SOTA on ImageNet for the past few years has been driven by pre-training on massive datasets. Vision transformers are increasingly more common and are extremely data-hungry.

Using your own link, SOTA is 2.6%. WER could be lower for the much smaller vocabulary used in a drive-thru.

I'm not sure what you mean by "that doesn't work in practice" re: using logits. Word-level confidence is pretty useful with GCP Speech-to-Text [1]

[1] https://cloud.google.com/speech-to-text/docs/word-confidence....

As it says right at the top of their README: "Evolution of Xamarin.Forms"

"The Lord of the Rings is of course a fundamentally religious and Catholic work, unconsciously so at first, but consciously in the revision."

The Letters of J.R.R. Tolkien #142

Interesting, I didn't know that NeRF was simply a feedforward network.

I hope that this research group can make more headway into training on CPUs, but I also would like to (naively) see less hyperbolic titles. This paper is not just particularly relevant to wide networks - it's only relevant to wide networks.

The original paper includes convolutional layer support in their future work & next steps. But it's not a foregone conclusion that the same speedup will occur.

The title is a bit misleading as this algorithm is for feedforward networks and doesn't yet support convolutional layers or any of the SOTA techniques for image classification... which is why GPUs reign supreme for training deep neural nets.

This is false. You can prototype a network on an M1 [1] and teacher-student models are a de facto standard for scaling down.

You can trivially run transfer-learning on an M1 to prototype and see if a particular backbone fits well to a small dataset, then kickoff training on some cloud instance with the larger dataset for a few days.

[1] https://blog.tensorflow.org/2020/11/accelerating-tensorflow-...

This is just patently false. Most of the folks I know that have an M1 are students that were saving up to upgrade from a much older computer and got the M1 Air. I can assure you that they don't have a decent amount of money.

Tensorflow has a branch that's optimized for metal with impressive performance. [1] It's fast enough to do transfer learning quickly on a large resnet, which is a common use-case for photo/video editing apps that have ML-powered workflows. It's best for everyone to do this locally: maintains privacy for the user and eliminates cloud costs for the developer.

Also, not everyone has an imagenet sized dataset. A lot of applied ML uses small networks where prototyping is doable on a local machine.

[1] https://blog.tensorflow.org/2020/11/accelerating-tensorflow-...

Is this paper on arxiv? This overview doesn't answer any critical questions. For example, it's easy to fill up 128 references and a reader shouldn't blindly trust a claim that, "The version of the paper we saw does also nod to several research efforts on reducing the size and computational costs of large language models, and on measuring the embedded bias of models."

If a key part of Google's claim is that the paper omits relevant research, an author should have simply posted their 128 references and openly asked what work was missing. This whole saga could be easily solved instead of being dragged out for clicks.