thewataccount
You just use a wildcard dns record, and then the webserver uses the host header to determine which subdomain was used (for webservers at least).
And pilots can land with a visual approach and still have ILS.
You would still have ADSB with your altitude, and I beleive heading and airspeed.
Because of the existing requirement to be spotting traffic and being over 1000ft away from other planes, they should be fine.
ATC should still have radar vectoring for their airspace.
I assume planes use more then just GPS, and use galileo and the other networks like consumer GPS do.
You can just use a yubikey unless I'm mistaken?
First you define a list of tokens - lets say 24 letters because that's easier.
They are a machine that takes an input sequence of tokens, does a deterministic series of matrix operations, and outputs what is a list of the probability of every token.
"learning" is just the process of setting some of the numbers inside of a matrix(s) used for some of the operations.
Notice that there's only a single "if" statement in their final code, and it's for evaluating the result's accuracy. All of the "logic" is from the result of these matrix operations.
One of the previous uses of twitter was "announcements". Government, corporate, emergency, community, software, etc.
The previous move to require logging in was already a massive blow to that, requiring a payment would decimate it was a use-case.
I don't see company's/governments putting announcements on a paid platform. You can't expect people to pay for twitter for announcements.
Even if it was the same seed, from my understanding Dalle3 would have to be just a further trained version of the same checkpoint to even resemble Dalle2's image. Like stable diffusion 1.4 vs 1.5 and 2.0 and 2.1 will make identifiably similar images, but 1.5 vs 2.1 vs SDXL won't look remotely similar.
Even more so because I'd wager they changed their encoder and/or decoder too.
* I think that if they generated something like a controlnet for guidance the same way in both models then they might be comparable but from my understanding Dalle2 doesn't work that way at all.
Comparisons would still be interesting though!
AFAIK every app that uses the ImageI/O api is effected by it, which includes every app you mentioned. You often don't need to even open the message for the image to be decoded.
From my understanding most vulnerabilities are from either the image decoder, text decoder, or webkit which again, effects nearly all apps. All apps can only use the webkit view, which affects nearly all of them to some degree.
I think you might be confusing the attack vector - messages is the easiest to attack since you just sent a regular text. Even if you don't normally use messages, it'll parse the image and you'll be hit by the 0 day. In theory this will work with most messaging apps.
Does anyone have any tips for creating sufficient datasets for finetuning specific workloads?
They still do security fixes for iOS 15 which is the iphone 6S in 2015 so 8 years.
The last security update for iOS 12 was 7 months ago too.
IMO that's a generous period for security updates.
Their usefulness is honestly more limited by app developer's support of the old api's, and the increasing bloat of web apps rather than Apple's support. It would be better of course if we could homebrew on them after that but I'm not sure I'd call it pitiful.
I see it's briefly mentioned in their docs under the deprecations section.
https://platform.openai.com/docs/deprecations
> Note: The recommended replacement, gpt-3.5-turbo-instruct, has not yet launched. Impacted customers will be notified by email when it becomes available.
lmql might be a decent alternative.
Any form of logit bias should work though.
I cannot recommend guidance enough. You can use shockingly small Llama models for some tasks with guidance while only actually generating a handful of tokens.
You should highly consider some form of guidance/logit bias for classification especially if you have a known set of classes. This will ensure you get it in the format that you want, with the correct classes that you want.
Keep in mind LLMs perform much better with COT. So you make it explain what the text/image is, then explain the possible classifications, then list its final decision. Again guidance can ensure it follows the correct format to do this.
LLM's still massively benefit from finetuning, especially if you want too classify it in a particular format. Notebook tags vs SFW/NFSW vs important subjects, etc. Existing alignment can sometimes mess with some of these classifications too which finetuning helps smooth out.
I do however highly doubt Cinder would be of use to Mastodon at the current moment.
What RAG here is doing is using embeddings and a vector store to identify close pieces of information, for example "in this django project add a textfield" will be very close to documentation in the django docs that say "textfield", and it will then add that to the prompt so the LLM has the relevant docs in its context.
The problem is that you'll need a heuristic to identify at least "potentially anomalous" and even then you'll still have to make sure there's enough context for it to know "is this a normal daily fluctuation".
A multi-step agent is definitely what you want, you could have it build an SQL query itself, for example "was there any high latency requests yesterday?" it may identify it should filter the time, possibly design the query to determine what is "high".
---
At the moment I don't think it's well suited to identifying when the "latency is abnormally high". However if you have some other system/human identify heuristics to feed to the LLM, it may then be able to do at least answer the query.
It's probably worth noting Meta made their own JIT'd version of CPython just for Instagram/Threads. Maybe mastodon devs can use their interpreter, but Meta's human/compute resources are magnitudes more than mastodon, the lessons from Meta might not be very applicable.
Google - T5 - Feb 2020 - https://blog.research.google/2020/02/exploring-transfer-lear...
Both of these were and still are used heavily for on-going research and T5 has been found to be decently useful when fine-tuned.
Weights were available for both.
My thing is that dynamically doing that is still a lot compared to just calling a single endpoint and all of that is handled for you.
But for sure this is a very decent horizontal use-case.
Hopefully more on-demand services enter the space. Currently where I am we don't have the resources for any type of self orchestration and our use case is so low/sporadic that we can't simply have a dedicated instance.
Last I saw the current services were rather expensive but I should recheck.
So you'll have to figure out how to run/scale the model inference. Cloud GPU instances are generally very expensive, and once you start needing to horizontally scale it'll get messy fast.
At least at the moment it's expensive, especially if it's either very light usage or very intensive usage - you either need just a few seconds of compute occasionally, or lots of compute all the time requiring scaling.
The "lucky" ones in this scenario are small-medium businesses that can use one or a few cards on-site for their traffic. Even then when you take the cost of an A100 + maintaining it, etc. OpenAI's offering still looks attractive.
I know there's a few services that try to provide an api similar to what openai has, and some software to self orchestrate it, I'm curious how those compare...
I will say though, from my experience (with other software) what usually happens is the original plugins/software get's abandoned or neglected so they can focus on the official implementation. And the official implementations rarely feature match everything from the original.
So hopefully that doesn't happen
Just choose a template immediately that has the creation date, and tags.
Tag your notes, just attempt to be consistent with the tags. Obsidian has some crazy powerful plugins that let you build it to your workflow - but if you start by tagging everything you'll be set moving forward.
I personally like the unique note creator to prepend dates.
Quickadd+Commander lets you make buttons for individual templates. Dataview is also very useful.
It took me two years before I really started personalizing it, I currently have templates for "regular notes", "weekly meeting", and "q/a notes". I'm adding notes to list my #wip as well
For me and I think a lot of others that's one of their main features over the competition.
I'm concerned about the current download's availability - its two URLs to some object storage. I find that these go dark rather quickly for many different reasons (accidentally moving it, bandwidth limits, deleting it later, etc).
I'm curious if there's a reason it's not also hosted on huggingface? I'm not saying they're the best place, but redundancy is good, most models have entries there, they have a very good cdn, and isn't as likely to go dark accidentally.
> Especially when it comes to prototyping
That makes sense! You're definitely right that github pages does have added friction for mono-repos and also the sporadic projects I have.
Asking seriously without any intention of being demeaning this looks like a great project, new alternatives are always welcome!
Doesn't git solve all of these issues other than "not creating the project first"?
For example github pages offers version control, custom domains, asset hosting, https, deployment with a similar type of command "git push"?