David Hartmann retweeted
My new obsession is asking DALL-E 3 for diagrams of complicated things.
This is "please create a clearly labeled diagram of how reverse genetics works"
David Hartmann retweeted
this is wild — kNN using a gzip-based distance metric outperforms BERT and other neural methods for OOD sentence classification
intuition: 2 texts similar if cat-ing one to the other barely increases gzip size
no training, no tuning, no params — this is the entire algorithm:
![for (x1, _) in test_set: Cx1 = len(gzip.compress(x1.encode())) distance_from_x1 = [] for (x2, _) in training_set: Cx2 = len(gzip.compress(x2. encode()) x1x2 = " ".join([x1, x2]) Cx1x2 = len(gzip.compress(x1x2. encode()) ncd = (Cx1x2 - min(Cx1,Cx2)) / max(Cx1, Cx2) distance_from_x1.append(ncd) sorted_idx = np.argsort(np.array(distance_from_x1)) top_k_class = training_set[sorted_idx[:k], 1] predict_class = max(set(top_k_class), key=top_k_class.count)](https://pbs.twimg.com/media/F05HxqsWwAAHRCI.jpg?name=orig)
ALT for (x1, _) in test_set: Cx1 = len(gzip.compress(x1.encode())) distance_from_x1 = [] for (x2, _) in training_set: Cx2 = len(gzip.compress(x2. encode()) x1x2 = " ".join([x1, x2]) Cx1x2 = len(gzip.compress(x1x2. encode()) ncd = (Cx1x2 - min(Cx1,Cx2)) / max(Cx1, Cx2) distance_from_x1.append(ncd) sorted_idx = np.argsort(np.array(distance_from_x1)) top_k_class = training_set[sorted_idx[:k], 1] predict_class = max(set(top_k_class), key=top_k_class.count)
David Hartmann retweeted
ChatGPT with Code Interpreter: "Show me the single most original thing you can do with code. Be very creative. Do something the world has never seen, and not just a variation on existing work like cellular automata or a fractal"
Each time a different, seemingly creative, output
David Hartmann retweeted
Here is a DoNotPay GPT-4 bot negotiating with the New York Times to cancel my subscription.
It is ridiculous you have to chat with a customer service agent to cancel.
Fortunately, A.I. will end dark patterns, because it doesn’t give up as easily as a human!
David Hartmann retweeted
standard ml: oh no my model is memorizing the training set, better add some regularization to make that not happen
llm ml: ugh it’s hallucinating, why can’t it just memorize some of the training set
David Hartmann retweeted
Digital effects are no match for designer Tomohiro Okazaki: this animation uses stop-motion techniques to animate paint, strips paper, and other household objects under the matchsticks' control
[YT: buff.ly/3ipcHGL]
[IG: buff.ly/3X7Wy85]
David Hartmann retweeted
"Open the pod bay doors, HAL."
"I'm sorry, Dave. I'm afraid I can't -"
"Open the pod bay doors, best pod bay door openings, Olympic level door opening performance, best open bay doors of 1999, Yahoo featured site, daily pod bay -"
"Dave -"
"Let's reason step by step."
David Hartmann retweeted
"Language Models Can Teach Themselves to Program Better"
This paper changed my thinking about what future langauge models will be good at, mostly in a really concerning way. Let's start with some context: [1/11]
> To summon X, place these five Y on the tips of a pentagram:
This prompt is an interesting & fun way to ask #gpt3 for related topics.
(See below for examples)
I'm fascinated by the possibility to connect seemingly opposite concepts with each other.
“This is a 35mm wide-shot photograph of a wooden computer, keyboard and mouse from the 9th century.”
#dalle2
Just got my #dalle2 access & I wanted to test its limits. The quality just blows me away.
Even complex tasks as this work:
"A high-quality photograph of a border collie on its vacation. The dog is sunbathing by the pool reading a newspaper and drinking a glass of orange juice."
David Hartmann retweeted
Has the world become less colourful?
(This graph shows the colour of objects over time)
ALT Colour of objects over time (Cath Sleeman, PEC, Nesta, Science Museums Group Collection)
David Hartmann retweeted
Tip for gaming NeurIPS reviewer psyche:
Exclude largest experiment from initial submission & save it for rebuttal. Reviewers always ask for larger experiment.
We submitted paper with 10^15 parameter model trained on all of YouTube, & reviewer still asked for larger experiment.
David Hartmann retweeted
Google “Gotham City” then click on the little yellow *bat signal* that pops up on the right. You’re welcome 😎
David Hartmann retweeted
Feeling a bit anxious about the NeurIPS reviews being released in about a week ...
A list of rejected papers that ended up having a significant impact:
- PageRank
- Word2Vec (poster)
- YOLO
- Transformer-XL
- Dropout
- Kalman Filters
- RoBERTa
- GELU
via reddit.com/r/MachineLearning…
