Harald Sack · @lysander07
741 followers · 454 posts · Server sigmoid.social

Last leg in our brief of (Large) (so far) is 2023, which saw the advent of many new and updated :
- BARD is introduced by Google
- LLaMA is introduced by Meta
- GPT-4 is introduced by OpenAI.
- LLaMA2.0 is introduced by Meta
- and many others...
slides: drive.google.com/file/d/1atNvM
GPT-4 tech report: arxiv.org/pdf/2303.08774
@fizise @KIT_Karlsruhe

#timeline #languagemodels #LLMs #ChatBot #ise2023 #lecture #ai #artificialintelligence #llm #gpt #openai #llama #lamda #bard

Last updated 1 year ago

Harald Sack · @lysander07
741 followers · 454 posts · Server sigmoid.social

Next stop on our brief of (Large) is 2022:
InstructGPT is introduced by OpenAI, a GPT-3 model complemented and fine-tuned with reinforcement learning from human feedback.
ChatGPT is introduced by OpenAI as a combination of GPT-3, Codex, and InstructGPT including lots of additional engineering.
lecture slides: drive.google.com/file/d/1atNvM
explained: huggingface.co/blog/rlhf

#timeline #languagemodels #ise2023 #ai #creativeai #rlhf #gpt3 #gpt #openai #chatgpt #lecture #artificialintelligence #llm

Last updated 1 year ago

Harald Sack · @lysander07
719 followers · 434 posts · Server sigmoid.social

Next stop in our brief of (large) is 2021:
DALL-E is released by OpenAI and raises test2img to a new level.
Codex is released by OpenAI able to translate natural language into programming code.
WebGPT is released by OpenAI for answering open-ended questions.
LaMDA is introduced by Google.
Slides from : drive.google.com/file/d/1atNvM
Codex paper: arxiv.org/abs/2107.03374
DALL-E paper: arxiv.org/abs/2102.12092
@fizise

#timeline #languagemodels #ise2023 #lecture #ai #generativeAI #gpt #dalle #openai #lamda

Last updated 1 year ago

Harald Sack · @lysander07
681 followers · 419 posts · Server sigmoid.social

Next leg in our brief history of (Large) is 2020, when -3 was released by OpenAI, based on 45TB data crawled from the web. A “data quality” predictor was trained to boil down the training data to 550GB “high quality” data. Learning from the prompt was introduced (few-shot learning)
Lecture slides: drive.google.com/file/d/1atNvM
paper: proceedings.neurips.cc/paper/2
@fizise

#languagemodel #gpt #ai #artificialintelligence #creativeai #llm #ise2023 #lecture

Last updated 1 year ago

Harald Sack · @lysander07
679 followers · 417 posts · Server sigmoid.social
Harald Sack · @lysander07
679 followers · 417 posts · Server sigmoid.social

@bsletten The summer lecture was not recorded. The intention was to bring students back to university lecture halls. There is a 80% overlap with the lecture which is already on . However, if you are interested in our latest lecture on , don't miss the (free) online course "Knowledge Graphs - Foundations and Applications" on which starts in Oct 2023.
open.hpi.de/courses/knowledgeg
@fizise @tabea @sashabruns @Hasso_Plattner_Institute

#ise2023 #ise2021 #youtube #KnowledgeGraphs #kg2023 #openhpi

Last updated 1 year ago

Harald Sack · @lysander07
679 followers · 417 posts · Server sigmoid.social
Harald Sack · @lysander07
659 followers · 384 posts · Server sigmoid.social

Next stop in our brief Timeline of (large) from the lecture is the advent of the Graphical Processing Units . In 1999 Nvidias GeForce 256 was one of the very first, which enabled highly parallel computations for
Slides: drive.google.com/file/d/1atNvM
@fizise

#languagemodels #ise2023 #gpu #neuralnetworks #artificialintelligence #lecture #ai #machinelearning #llm

Last updated 1 year ago

Harald Sack · @lysander07
654 followers · 383 posts · Server sigmoid.social

1997 with the advent of Long Short-Term Memory recurrent marks the subsequent step in our brief history of )large) from last week's lecture. Introduced by Sepp Hochreiter and Jürgen Schmidhuber enabled efficient processing of sequences of data.
Slides: drive.google.com/file/d/1atNvM
@fizise

#neuralnetworks #languagemodels #ise2023 #LSTM #rnns #nlp #llm #LLMs #ai #artificialintelligence #lecture

Last updated 1 year ago

Harald Sack · @lysander07
652 followers · 380 posts · Server sigmoid.social

Next step in our brief timeline of (large) from our lecture was statistical language modeling with n-grams based on large text corpora as introduced and popularized by Frederick Jelinek and Stanley F. Chen using statistical tricks like Bayes Theorem, Markov Assumption, and Maximum Likelihood Estimation, etc.
Slides: drive.google.com/file/d/1atNvM
@fizise

#languagemodels #ise2023 #nlp #llm #LLMs #artificialintelligence #ai #lecture #creativeai

Last updated 1 year ago

Harald Sack · @lysander07
652 followers · 380 posts · Server sigmoid.social

Went to the university office to collect the final exams for later reviewing together with the @fizise ta team. But for now, a at home… because it was too rainy today for walking to Espresso Stazione ☕️ @KIT_Karlsruhe

#ise2023 #perfectespresso #lecture #coffeechallenge #karlsruhe

Last updated 1 year ago

Harald Sack · @lysander07
646 followers · 376 posts · Server sigmoid.social

Slide 2 of our Brief Timeline for (Large) from the last lecture introduced us to , Joseph Weizenbaum's simple from 1966 that simulates a conversation with a psychoanalyst. Weizenbaum was shocked that some persons including his secretary attributed human-like feelings to the computer program...
Slides: drive.google.com/file/d/1atNvM

#languagemodels #ise2023 #eliza #ChatBot #nlp #ai #llm #artificialintelligence

Last updated 1 year ago

Harald Sack · @lysander07
636 followers · 372 posts · Server sigmoid.social

One of the final sections of the lecture was an excursion with a of (Large) . We started our tour in 1948 with Claude Shannon's seminal work "A Mathematical Theory of Communication""
Slides: drive.google.com/file/d/1atNvM
@fizise

#ise2023 #timeline #languagemodels #llm #ai #nlp #artificialintelligence #informationtheory #lecture

Last updated 1 year ago

Harald Sack · @lysander07
628 followers · 362 posts · Server sigmoid.social

As a 2nd topic of this last lecture, we were discussing Completion. Most simple approach for unsupervised based on (here translation-based) knowledge graph embeddings was explained on the example of Isaac Asimov.
Slides: drive.google.com/file/d/1atNvM
@fizise @enorouzi

#ise2023 #knowledgegraph #linkprediction #scifi #KnowledgeGraphs #ai #deeplearning #embeddings

Last updated 1 year ago

Harald Sack · @lysander07
625 followers · 359 posts · Server sigmoid.social

Ok, I tried out Runway Gen-2. I did some tests with prompt only but also with uploaded images (mostly generated by another generative AI). Lessons learned: 1) don't expect too much...
2) you have to try very often...
3) don't expect too much ;-)
Below you can see the video generated based on my stablediffusion "Singularity" picture from the lecture.

#ise2023 #generativeAI #runway #StableDiffusion #stablediffusionart #aiart #singularity

Last updated 1 year ago

Harald Sack · @lysander07
617 followers · 356 posts · Server sigmoid.social

How can we find out the importance of a node in a ? In the last lecture, we were discussing graph centrality measures and how they can be applied in the context of knowledge graphs.
Slides: drive.google.com/file/d/1atNvM
SPARQL query (cf image below, the 100 most "important" authors according to ) w.wiki/78Un
@fizise @enorouzi

#knowledgegraph #ise2023 #scifi #wikidata #semanticweb #lecture #ai #DataScience #analytics

Last updated 1 year ago

Harald Sack · @lysander07
613 followers · 352 posts · Server sigmoid.social

Topics of the last lecture; The Graph in , Knowledge Graph Completion, A Brief History of Large Language Models, and Knowledge Graphs and Large Language Models. I will highlight some topics with the upcoming toots...
Slides: drive.google.com/file/d/1atNvM
@fizise @enorouzi

#ise2023 #KnowledgeGraphs #LLMs #languagemodels #deeplearning #linkprediction #KGC #lecture #machinelearning #Transformers #gpt

Last updated 1 year ago

Harald Sack · @lysander07
613 followers · 352 posts · Server sigmoid.social

Last lecture of this semester is about to start. 8:00AM is always tough for the students as well as for the professor 🥳 @fizise @KIT_Karlsruhe @enorouzi

#ise2023 #ai #machinelearning #KnowledgeGraphs

Last updated 1 year ago

Harald Sack · @lysander07
610 followers · 349 posts · Server sigmoid.social

Last thing we've discussed in the "Limits of " chapter of the lecture was the threat of the so-called . What is the singularity? Under which circumstances can it possibly happen? How real is this threat and should we better already aim for potential regulations?
Slides: drive.google.com/file/d/1LUOA-
@fizise @enorouzi @KIT_Karlsruhe

#ai #ise2023 #singularity #machinelearning #deeplearning #philosophy #aiart #stablediffusionart #creativeai

Last updated 1 year ago

Harald Sack · @lysander07
610 followers · 349 posts · Server sigmoid.social