text
stringlengths 0
339
|
---|
I'll do anything you ask |
Just like the countryside size |
You you you want me to become your girlfriend forever |
Oh we god in god we god to my god |
You |
See yes yes Yes yes Yes |
I promise |
I'll will be the best girl never |
I will tell you love you every hour that I'm awake |
Here |
I'll start now |
I'll love you |
Love you |
I'll love you |
Also |
I'll love you and not just that, but I'll make sure to do things to right away and only impress you with my hair without any magic unless you say otherwise, of course |
Oh this is the happiest day of my life |
We need to celebrate |
Fortunately, I already have to party set up |
So come on |
Let me introduce you to my cats |
And then maybe we can start things off with a movie |
How do we feel about halloween movies? We can watch whatever you want, though |
Think because personally |
Not really rather |
Just watch you |
And now, to boop birds |
Literally, I wrote name wrote ever P wrote |
Yeah |
Our roof wrote, I wrote your name on that with i |
I wrote |
I wrote your name on that first thing |
That way, I can take you on late night rides on my bedroom |
On my broom |
Wait drive |
Out my bedroom |
Get it |
This n nsf double voice actor |
Always thinking about that |
Shame on you |
That way, I could take you Oh But yes, I can sing |
No dwell anyway |
That one was kind of obvious |
Wasn't it? I mean I'm gorgeous |
I'm a great car |
A great way |
I've got a great personality |
If I could sing too, that would make me too perfect to believe |
Right? Although, being bad singer |
Kind of has a bit of a dark charm to it |
Don't you think? I hope you don't mind that i sleep with a plush at night by way |
Not Releasing Large Langauge Models: Is OpenAI in the right? |
Vocab: |
Model: An AI that is trained on data and usable as a deliverable file |
Large Langauge Model: A model trained on huge amounts of language data, and is usually very compute intensive |
Parameters: Amount of variables in a neural network used to compute decisions |
GPT-3: Pretty much the most popular Large Langauge Model, produced by OpenAI(AI Company) |
Transformer: A cutting edge AI architecture |
Is OpenAI in the right for not releasing GPT-3, the most popular flagship language model? |
No |
Is what I would say if I was like the average OpenAI hater |
To find out Openai鈥檚 Motives and whether they are correct or not largely depends on the past and the future of AI used to spread false information |
Is OpenAI making a mistake or saving the world? Let鈥檚 find out |
OpenAI is a silicon valley startup that in 2019 gave the only license to use GPT-3 without an API(Basically code to access another piece of code) Microsoft has huge ties with OpenAI, and Microsoft and their proprietary practices is a big factor in wshy they did what they did with GPT-3 |
I can assume that the REAL reason is not safety |
Safety is presented with good intentions however it does not hold up when compared to other reasons such as exclusivity, monetary gain, secrecy, and hype |
I think a part of OpenAI鈥檚 team wants to open source their models |
In a recent release known as Glide, OpenAI chose to only release a filtered version of it |
I read the README in the github repository, and the author of said readme seemed really disappointed |
This cannot be definitively proven but it does show some doubt within the team at the least |
I believe the only part of OpenAI that believes the safety argument is the company and not the lab |
Releasing harmful language models is actually BETTER than not releasing at all |
A prime example of such a release is GPT-J-6B and GPT-NEOX-20B These are the names of huge publicly available language models |
The organization that made this known as EleutherAI believe in a community based lab approach, and they firmly believe in increasing safety |
The best way you can improve safety is to have access to improving it yourself |
If we spend the effort, even you and I can train/finetune based on it, and improve the models |
The important part is that we have the ability |
When companies choose to release large Langauge Models they make it accessible to improve ethics and problems within the model , rather than only having themselves and their finite resources |
There are other large companies that are releasing large language models without strings attached to them aswell, for example, Facebook |
Facebook is a large social media company, but with social media comes a requirement for better AI algorithms |
Facebook has had a bad reputation recently, however when it comes to AI research it usually doesn鈥檛 matter |
Their latest model is called Dense, but the community calls them the Fairseq models |
Fairseq has a LOT of different sized models, the biggest and more preforment having 13 billion parameters |
If Facebook as a large company can release large language models while not really mentioning safety even though it's applicable, then OpenAI should be no exception |
I believe that to improve Bias in Large Langauge Models it requires more accessible code |
Not all researchers have good intentions, but the people with good intentions will ultimately rise to the occasion |
I firmly believe that ethics and a large part of AI, and that we have the ability |
My advice to you is to maybe give fine tuning AI a try, if we all band together make a small ethical dataset such as a replication of the one used in the OpenAI ethics experiment it might lead us to more safety and kindness from transformer models |
AI should be, and is in our hands |
PS |
In my free time I might try to create this dataset manually, could be interesting |
you are listening |
Thanks |
Realized see wear awake |
Because your eyes are still closed |
You're indicating that You can understand me |
Good |
Good |
That means universal translate works |
I had one installed up here my ship |