I'm in a new building that uses sound isolating tech in the US and it's fantastic. You get some noise coming in from the front door since it's metal but both neighbors around me have dogs and I've almost never heard them. Every once in a while I can hear the slight vibration of the dog upstairs dropping its ball on the ground, but that's about it.
Not that they'd do this, but if the company offered what it was paying for office space for you as a stipend for your home office, would that change your mind? Some quick Google math suggests an average of $80/sqft/year for NYC office space and suggests 100 sqft per-person for an open-office worker which comes out to $600-700 a month. Again, cursory Google search, but it looks like that's right about the price increase to go from a 2BR to a 3BR and have a home office.
At the most basic, you give it text and it can guess at what comes next. This means you could type "10 types of ferns" and it will build a list of 10 ferns. Or you could type it out how a transcript of a conversation would look like and it will basically fill in the other "side" of the conversation to make a chatbot (all the complicated chatbots are basically abstracting this). Think of it like a text box with a super-smart (arguably) person also having access where you can type one thing and then it'll type whatever it thinks would be the next thing to type.
This seems incentivized by marketing, as every security demo I've ever had has included something like "See? Our tool found 46,291 possibly critical issues in your environment" and then you drill into them later and they're relatively insignificant issues you'd fix in a perfect world with unlimited labor but drown out the actual things you actually want to know about right now. I think there's also incentive to upgrade the threat urgency on alerts to look like they're capturing more urgent threats.
> My biggest recommendation for selling/pitching/positioning anything in cyber is to solve problems I already am dealing with. Do not go to a customer with a flashlight that points out new, unknown issues and requires additional resources to fix... it's fine to point out problems, so long as you have an easy button I can smash to fix it, without having to pull in someone to deal with it.
I hear this one. I can't count the amount of times some salesperson has managed to get the ear of a decision-maker and try to insist that we need to solve some new urgent problem category that the vendor's sales department has invented. The solution always promises to solve all our "newly discovered" problems but the solution requires us to dedicate someone 20-40 hours a week to just working their platform. Another fun one is they advertise that it's "automated" but the setup is so long and arduous with required organizational process changes that break workflows that it sits there doing 5% of the advertised capabilities like a gym membership you don't use often, but feel like you've admitted defeat if you cancel it.
Could you give us a bit of what it was like? I imagine it being somewhat like the Vegas hotels where you can spend an entire vacation without actually having to step outside or a military/cruise ship that has stores/entertainment/exercise facilities built right in.
You are exactly right. And to address the concerns about errors made by the machine, we are not looking in my organization to have the machine automatically make decisions on things like access. It would ideally warn us when we've likely done or are about to do something wrong. An auditing tool at least.
Just be careful about paying for the (one time fee) Pro Apollo at this point. Reddit just changed their API terms and that app may stop working soon-ish without a monthly subscription.
Once this barrier is broken down we'll see a lot of cool things. 32k on GPT-4 is already pretty cool but once we get into hundreds of thousands/millions of tokens of context we'll be able to easily do things only currently achievable with fine tuning and "memory" tricks. Assistants that remember everything you've ever told them, asking detailed questions about large datasets, even complex systems that are bootstrapped from the context.
I think a real promise with these smaller models is in upgrading/replacing Siri/Alexa in mobile devices, especially when it's trained on your messages/habits/calendar/notes/shopping/music. A true virtual assistant you can say "I want to have a karaoke party, can you plan one for next Saturday?" and it creates a guest list, invites your friends it knows like karaoke (even the ones you forgot about!), puts it on your calendar, reminds your friends, sets up a delivery of the party supplies based on what it knows you have on-hand, handles rsvps, looks into pricing of having your place professionally cleaned, and creates a curated playlist. It might not be able to do all these things with its local capabilities, but it can know when to reach out to more powerful models to execute the more challenging tasks.
You're not lying. When I saw the site I thought "isn't that the dude who managed to emulate an entire PC in Javascript and boot Windows in a browser? Yep that's him."
Quality and quantity: I've had the best luck cramming a bunch of examples into the input, just like with GPT-J where you're only working with 6B parameters. Making sure the format stays consistent and ideally presented in the shape you'd encounter that same text if you found it on a webpage somewhere.
If so, it sounds like there's even less reason for the FDA to be involved if the DEA is already catching the truly illicit stuff. The issue the article takes is that the FDA keeps asking for more and more money to inspect packages for opioids/fentanyl, but is using that money to mostly destroy legitimate medicine. If they had to say what they were actually doing with the money, they might not get it.
I'm astonished that they treat those alarms like I treated the check-engine light as a poor kid who reasoned that the car seems to still be running fine.
They're claiming the 13b model beats GPT-3 175b which is an extraordinary claim requiring extraordinary evidence. If that's true, though, it'd be interesting to see if that also applies to fine-tuning. Since the claim is predicated on the 13b model being better trained (amongst other things?), I wonder if limited fine-tuning data handicaps the 13b model even if the base model can outperform GPT-3 Davinci, given your point about large models handling fine-tuning better with limited data.
This is cool! But I wonder if it's economical using cloud hardware. The author claims 1.12 tokens/second on the 175B parameter model (arguably comparable to GPT-3 Davinci). That's about 100k tokens a day on the GCP machine the author used. Someone double check my numbers here, but given the Davinci base cost of $0.02 per 1k tokens and GCP cost for the hardware listed "NVIIDA T4 (16GB) instance on GCP with 208GB of DRAM and 1.5TB of SSD" coming up to about $434 on spot instance pricing, you could simply use the OpenAI API and generate about 723k tokens a day for the same price as running the spot instance (which could go offline at any point due to it being a spot instance).
Running the fine-tuned versions of OpenAI models are approximately 6x more expensive per token. If you were running a fine-tuned model on local commodity hardware, the economies would start to tilt in favor of doing something like this if the load was predictable and relatively constant.
I was curious about this before, assuming that it must be a far cheaper way to live. Turned out that after lot rental fees and the fact that mobile homes depreciate in value rather than appreciate, the economics just don't seem to work out unless your largest requirement is to be in a physically separate building than your neighbors vs being in an apartment.
Whole systems from a single prompt are probably a ways away, but I was able to get further than I expected by asking it what classes would makeup the task I was trying to do and then having it write those classes.
Mixed-use high-rises are really cool. You can go to the mall, the gym, the pool, see a movie, go to the dentist, or go grocery shopping just by taking the elevator. They almost feel like resorts you can live in.
I wonder what the transit modes look like on a macro scale for those who move downtown. Anecdotally, it seems common for those who move to downtown Chicago to get rid of their car as a spot in a safe area can easily run $300 a month, and with a payment, maintenance, and insurance it really adds up. Being so close to all the major train lines, I myself have switched entirely to public/shared transit.
I've done outsourced IT work for a bunch of small companies. What I've seen so often was something like "we had a very talented programming intern who built this whole thing and then (went back to college | got a new job | accidentally deleted the codebase). So this time we went with an agency, but they called all the code garbage and said we need a complete re-write so now we have two systems we're moving between".
It seems on the other end that many companies are realizing the only thing cheaper than an empty room with some Ikea desks shoved side-to-side is to have people work remotely and pay for their own office space.