#NLP
BigDataLove.com #AI »The Humane Ai Pin finally debuts — on the catwalk at Paris Fashion Week« https://www.theverge.com/2023/9/30/23897065/humane-ai-pin-coperni-paris-fashion-week #BigDataLove #BigData #ML #NLP
Latest newsletter - image artifacts and claims of authenticity, 3d gaussian splats, testing DALLE3, an artist using a custom model, nlp tutorial notebooks, the game Control, tons of book recs (AI & ghosts). https://open.substack.com/pub/arnicas/p/titaa-47-authenticity-and-control?r=sv8a&utm_campaign=post&utm_medium=web #books #sff #aiart #3d #nlp
I don't want to explain the ethics of #AI but let me explain it. OpenAI does work but it doesn't do what people think it does.
What it does is actually making things sound more human. It doesn't generate it's own answer at all. In fact it is directly plagiarizing fed data. Google's Bard is more transparent in this aspect.
That is what #NLP is. This isn't some thinking machine. It's just something that takes internet posts and makes it sound better.
Grupo de Brasileiras com o foco em PLN. É um lindo projeto. Merece muito ser divulgado. Parabéns as 150 pesquisadoras 👏👏👏.
BigDataLove.com #AI »Who will benefit from AI?« https://news.mit.edu/2023/who-will-benefit-ai-machine-usefulness-0929 #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »AMD CEO Lisa Su on the AI revolution and competing with Nvidia« https://www.theverge.com/23894647/amd-ceo-lisa-su-ai-chips-nvidia-supply-chain-interview-decoder #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »The synthetic social network is coming« https://www.theverge.com/2023/9/29/23895675/ai-bot-social-network-openai-meta-chatbots #BigDataLove #BigData #ML #NLP
Our study delves into model editing utilizing in-context learning, aiming to improve LLM responses without the need for fine-tuning or retraining. Specifically, we propose a comprehensive retrieval strategy to extract medical facts from an external knowledge base, and then we incorporate them into the query prompt for the LLM.
BigDataLove.com #AI »Spotify is adding auto-generated transcripts to millions of podcasts« https://www.theverge.com/2023/9/28/23894674/spotify-auto-generated-podcast-transcripts #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Google adds a switch for publishers to opt out of becoming AI training data « https://www.theverge.com/2023/9/28/23894779/google-ai-extended-training-data-toggle-bard-vertex #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Newsrooms around the world are using AI to optimize work, despite concerns about bias and accuracy« https://www.theverge.com/2023/9/28/23894651/ai-newsroom-journalism-study-automation-bias #BigDataLove #BigData #ML #NLP
Job opportunity at CLASP in the Faculty of Humanities at the University of Gothenburg. Postdoc in multimodal coreference, two years, applications due Nov 21.
More about MESLaM, the language model I built out of matchboxes! In this post, I explain how I trained 35 matchboxes to generate novel text.
https://dcorney.com/thoughts/2023/09/26/training-matchboxes.html
It's machine learning, but the only machine is a sharp pair of scissors.
#NLProc #NLP #GenerativeAI
BigDataLove.com #AI »Mark Zuckerberg can’t quit the metaverse« https://www.theverge.com/23894188/mark-zuckerberg-metaverse-connect-2023-ai-vr #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Details emerge on Jony Ive and OpenAI’s plan to build the ‘iPhone of artificial intelligence’« https://www.theverge.com/2023/9/28/23893939/jony-ive-openai-sam-altman-iphone-of-artificial-intelligence-device #BigDataLove #BigData #ML #NLP
In this paper, we first present a multi-modal clinical conversation summary generation task that takes a clinician-patient interaction (both textual and visual information) and generates a succinct synopsis of the conversation. We leverage an adapter to infuse knowledge and visual features.
In "LLM Prompt Engineering For Developers," we embark on an in-depth exploration into LLMs and the art of constructing helpful prompts. The guide begins with a look at the history of Natural Language Processing (NLP) from its beginnings to present day. You will gain a thorough understanding of GPT models and their components, including architecture, capabilities, and intricacies https://leanpub.com/LLM-Prompt-Engineering-For-Developers #books #LLMs #AI #NLP
BigDataLove.com #AI »Adobe’s full Photoshop on the web launch includes its popular desktop AI tools« https://www.theverge.com/2023/9/27/23892889/adobe-photoshop-for-the-web-firefly-ai-generative-fill-full-release-price-date #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Big Data & AI In Collision Course With IP Laws – A Complete Guide« https://www.smartdatacollective.com/big-data-ai-in-collision-course-with-ip-laws-a-complete-guide/ #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »4 Ways AI Can Enhance Your Marketing Strategies« https://www.smartdatacollective.com/ways-ai-can-enhance-marketing-strategies/ #BigDataLove #BigData #ML #NLP
@mistralai just shared their 7B LLM with some neat trick to reduce computing needs and interesting performances.
https://mistral.ai/news/announcing-mistral-7b/
Nothing on the training data and process, though... this trend is quite deceiving. The HF model card is almost empty.
BigDataLove.com #AI »Microsoft is testing a DALL-E-powered text-to-image creator in Paint for Windows 11 « https://www.theverge.com/2023/9/27/23892900/microsoft-dall-e-windows-11-paint-cocreator #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »ChatGPT can now search the web in real time« https://www.theverge.com/2023/9/27/23892781/openai-chatgpt-live-web-results-browse-with-bing #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Re-imagining the opera of the future « https://news.mit.edu/2023/re-imagining-opera-of-the-future-valis-0927 #BigDataLove #BigData #ML #NLP
This month in Acuitas work, I'm still pushing on three fronts (TextGen, TextParse, and Narrative) but TG is finished and I do have a few little results to show.
https://writerofminds.blogspot.com/2023/09/acuitas-diary-64-september-2023.html
BigDataLove.com #AI »Artifact is becoming Twitter, too« https://www.theverge.com/2023/9/27/23887416/artifact-mike-krieger-code-2023-posts #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Mark Zuckerberg on Threads, the future of AI, and Quest 3« https://www.theverge.com/23889057/mark-zuckerberg-meta-ai-elon-musk-threads-quest-interview-decoder #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Facebook and Instagram will soon get a slew of AI-powered creator tools« https://www.theverge.com/2023/9/27/23892231/meta-generative-ai-image-editing-stickers-instagram-facebook-whatsapp #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Meta is putting AI chatbots everywhere« https://www.theverge.com/2023/9/27/23891128/meta-ai-assistant-characters-whatsapp-instagram-connect #BigDataLove #BigData #ML #NLP
I used #nlp and #linguistics (but no #genai #generativeAI) to generate more phrases like "champagne for my real friends, real pain for my sham friends."
Read up thread for more...
And if you have a good Python near-rhyme-detection implementation (or hints on how the phonology works), please let me know!
BigDataLove.com #AI » From physics to generative AI: An AI model for advanced pattern generation« https://news.mit.edu/2023/physics-generative-ai-ai-model-advanced-pattern-generation-0927 #BigDataLove #BigData #ML #NLP
Nuovo articolo (in italiano) pubblicato su Sistemi Intelligenti con @tommaso_caselli Malvina Nissim, Viviana Patti:
"Sono solo parole. ChatGPT: anatomia e raccomandazioni per l'uso" https://www.rivisteweb.it/doi/10.1422/108131
BigDataLove.com #AI »How TikTok Has Exposed Celebrities And Politicians’ Closest Personal Contacts« https://www.forbes.com/sites/alexandralevine/2023/09/27/tiktok-exposed-politicians-celebrities-contacts-privacy-security-election-china/ #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Jony Ive is reportedly developing an AI gadget with OpenAI’s Sam Altman« https://www.theverge.com/2023/9/27/23892135/jony-ive-sam-altman-ai-device-openai-ceo-apple-designer #BigDataLove #BigData #ML #NLP
Using AI to code? Be careful about this nlp #tech, more at https://maolte.ie/blog/18usingai4code.html #technology #programming #artificalintelligence #nlp #code
I asked Google's Bard: Who was wearing the pajamas in the sentence "While hunting in Africa, I shot an elephant in my pajamas"? So! Because I own the pajamas only I can wear them?
Response: ...the speaker was wearing the pajamas... the speaker uses the possessive pronoun "my" to describe the pajamas, which indicates that they belong to the speaker. However, it is important to note that this is a fictional sentence, and it is not recommended to hunt elephants in your pajamas.
#AI #nlp
BigDataLove.com #AI »The new WGA contract will change how Hollywood works« https://www.theverge.com/2023/9/26/23891835/wga-contract-summary-ai-streaming-data #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Eufy’s latest security package can track one person across multiple cameras« https://www.theverge.com/2023/9/26/23890980/eufy-security-cross-camera-tracking-ai-solocam-floodlight-doorbell #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »Ceramic Film Memory & New Spintronic Memory And Logic Materials« https://www.forbes.com/sites/tomcoughlin/2023/09/26/ceramic-film-memory--new-spintronic-memory-and-logic-materials/ #BigDataLove #BigData #ML #NLP
BigDataLove.com #AI »The Creator retreads familiar AI panic territory to stunningly inert effect« https://www.theverge.com/23889089/the-creator-movie-review-gareth-edwards #BigDataLove #BigData #ML #NLP
Our study analyzes germline abstracts in the construction of knowledge graphs of the of the immense work that has been done in this area for genes and diseases. This paper presents SimpleGermKG, an automatic knowledge graph construction approach that connects germline genes and diseases.
In this work we propose a hybrid model leveraging LayoutLMv3 along with domain-specific rules to identify key patient, physician, and exam-related entities in faxed referral documents. We explore some of the challenges in applying a document understanding model to referrals, which have formats varying by medical practice.
In this work, we investigate the ability of state-of-the-art large language models (LLMs) on the task of biomedical abstract simplification, using the publicly available dataset for plain language adaptation of biomedical abstracts (PLABA).
I'm pleased to reveal a new low-tech small language model called MESLaM! It required no GPUs to train, the training data is only 800 words and it's made entirely of matchboxes and paper -- but it's perfectly capable of generating novel text!
https://dcorney.com/thoughts/2023/09/23/matchboxes.html
Further evidence, if needed, that language models large or small are not intelligent.
#NLProc #NLP #GenerativeAI #LLM
Welcome the new babies!
19 pretrained models on the loose track
24 on the strict
118 on strict-small
https://dynabench.org/babylm
We are proud of >30 pretraining teams submitting papers to babyLM!
FOMO?
Get updated on CoNLL or
participate next year
https://babylm.github.io
#NLP #nlproc #babyLM #CoNLL #machinelearning #llm #llms #pretraining
Should healthcare insitututions train an LLM from scratch or fine-tune from an open-source model? And what are the biggest LLMs they can afford? In this study, we take steps towards answering these questions with an analysis on dataset sizes, model sizes, and costs for LLM training using EHR data.
This paper is focused on answering health-related questions asked by the public by providing visual answers from medical videos. We created two large-scale datasets: HealthVidQA-CRF and HealthVidQA-Prompt. We then proposed monomodal and multimodal approaches that can effectively provide visual answers from medical videos to natural language questions.
In this paper, inspired by Chain-of-thought, we leverage the LLM to solve the Biomedical NER step-by-step: break down the NER task into entity span extraction and entity type determination. Additionally, for entity type determination, we inject entity knowledge to address the problem that LLM's lack of domain knowledge when predicting entity category.
4/end
REFERENCE
[1] Ali Mousavi, Xin Zhan, He Bai, Peng Shi, Theo Rekatsinas, Benjamin Han, Yunyao Li, Jeff Pound, Josh Susskind, Natalie Schluter, Ihab Ilyas, and Navdeep Jaitly. 2023. Construction of Paired Knowledge Graph-Text Datasets Informed by Cyclic Evaluation. http://arxiv.org/abs/2309.11669
3/
* For out-of-distribution evaluation, we showed LAGRANGE-trained models, when subject to two separate cyclic evaluations w/ TeKGen and LAGRANGE, outperform the models trained with the other automatically constructed datasets, even the one for which train and test distributions match (TeKGen). The consistency in model rankings between these two cyclic evaluations also confirm the stability of the evaluation methodology (screenshot).
2/
* We constructed a large graph-aligned dataset called LAGRANGE (screenshot 1), and showed when used to train a T5-large model it performs the best among the automatically constructed datasets, even outperforming the manually constructed dataset WebNLG on BLUE-4 and ROUGE-4, and outperforming LLM-generated (#ChatGPT and #Guanaco ) datasets on graph construction metrics (screenshot 2 & 3).
1/ Our paper on automatic #knowledgeGraph-aligned dataset construction is out! The main points [1]:
* We showed cyclic evaluation — the process of training GTG (graph-to-text-to-graph) or TGT using a graph-aligned dataset (screenshot 1) — reflects faithfully the same trend a unidirectional evaluation does (screenshot 2-3). It is therefore a better way to assess data quality because it does not rely on knowing ground truth matches!
@emollick This has been an interesting experiment... I gave #AI #NLP #LLM #GPT4 custom instructions to use two files, a memory.txt file and an ideas.txt file, and asked it to record interesting ideas and maintain a memory of our interactions. You can see the results after a short session, it works better than I thought!
Custom prompt:
You should have access to two text files, Memory.txt and Index.txt. If you do not have the files, then ask me for it.
I'm finishing up a fun computer -assisted wordplay project (no generative AI involved, just old-school natural language processing #nlp #linguistics) and I can't wait to share it with y'all.
super interesting new paper showing that LLMs don't generalize A = B to B = A:
Negate v1.1.1 released! 🥳
https://github.com/dmlls/negate/releases/tag/v1.1.1
This new update implements inversion negation, which means that questions can now also be negated.
Curious how Negate is being used in research? Then read our paper at https://arxiv.org/abs/2307.13989.
Looking for a postdoc? I did too.
Lost? I'm sharing the tips i've learnt here:
https://borgr.github.io/post/
Do you have any questions? tips?
Tell me and let's improve it together for everyone.
We present Clinical Prediction with Large Language Models (CPLLM), a method that involves fine-tuning a pre-trained Large Language Model (LLM) for clinical disease prediction, the task of predicting whether patients will be diagnosed with a target disease during their next visit or in the subsequent diagnosis, leveraging their historical diagnosis records.
We propose a novel framework that models dialogues between patients and healthcare professionals using AMR graphs, where the neural networks incorporate textual and graphical knowledge with a dual attention mechanism. Experimental results show that our framework outperforms strong baseline models in medical dialogue generation.
This research article introduces a methodology for generating translated versions of annotated datasets through crosslingual annotation projection. Leveraging a language agnostic BERT-based approach, it is an efficient solution to increase low-resource corpora with few human efforts and by only using already available open data resources.
We propose a simple but effective model called the Multi-Hop Label-wise ATtention (MHLAT), in which multi-hop label-wise attention is deployed to get more precise and informative representations. Extensive experiments on three benchmark MIMIC datasets indicate that our method achieves significantly better or competitive performance on all seven metrics.
My favorite AI forum, https://aidreams.co.uk/ could really use some more members. It's open to enthusiasts, hobbyists, and/or professionals and enables long-form discussions that are harder to do on social media. It's also not dominated by people working on ML/transformers, so if you've found other AI communities a bit ... lacking in variety ... it's a good place for you. Come check it out!
Is there anybody in my circles working with/in #digitalHumanities? I am currently exploring possibilities of a PhD project leveraging digital technologies in #Persian language materials.
#iran #academia #digital #phd #protest #poetry #nlp #NaturalLanguageProcessing #helpWanted #GettingStarted
Sean Haas of the Advent of Computing podcast ought to get more attention around here! He's 116 episodes into a unique and rather excellent history of computing. He doesn't cover the mainline of computer history, but instead follows fascinating threads of his own. He tends to avoid the political implications of the work he describes, focusing instead on technical and administrative structures.
I'm thoroughly enjoying his two-parter on the history of #Prolog.
I liked his outline of 'first principles' languages, all of them #homoiconic: #Lisp, which operates on lists; #Forth, a tiny stack language; and #Prolog, a declarative logic language. The first and last of those were both originated for natural language processing.
https://adventofcomputing.com/
#PodcastPsaturday #retrocomputing #adventofcomp #AdventOfComputing #ProgrammingHistory #QuirkyHistory #OriginalResearch #NLP
🤖: Detecting if chatGPT made this text...
It did not
A survey on the (few) datasets and methods to detect it
https://arxiv.org/abs/2309.07689
(not sure why chatGPT and not LLM in general, but NVM)
#enough2skim #NLP #nlproc #chatgpt #LLM #LLMs #AGI
New #languagemodeling #nlp #ai #paper, led by Angelica Chen! We break the steepest MLM training loss drop into *2* phase changes: first in internal grammatical structure, then external capabilities. Big implications for emergence, simplicity bias, and interpretability! https://arxiv.org/abs/2309.07311
I am back from the 'Learning from Small Data' conference in Gothenburg. It was an excellent event which demonstrated once more that you need algorithmic diversity when dealing with the harder problems of #NLP. Transformers are not all you need. The question is, how do we give our community the time and space to experiment with more theoretically-grounded proposals when there is so much pressure to follow the trend?
I've used MeCab and UniDic for parsing Japanese text for years and a kind correspondent shared Ichi.Moe (e.g., https://ichi.moe/cl/qr/?q=ブラックシャドー団は集団で盗みを行う窃盗団でお金持ちの家を狙い、家にある物全て根こそぎ盗んでいきます%E3%80%82&r=htr) which uses Ichiran, a dictionary-only parser: it generates readings, tags parts-of-speech, and of course links dictionary entries!
This list of grammar that it handles https://github.com/tshatrov/ichiran/blob/99992db22e63287d3055e8539ad3b41665355457/dict-grammar.lisp#L110-L154 coincides very closely with what I've cobbled together with MeCab + dictionary + Kamiya-Codec (https://github.com/fasiha/kamiya-codec via https://github.com/fasiha/curtiz-japanese-nlp).
I don't think I'll be able to replace MeCab's machine learning-based parsing with Ichiran because MeCab handles Kansai-ben and colloquial speech pretty well which I don't think Ichiran can, but I definitely think combining MeCab with Ichiran could unlock a LOT of value. #ichiran #Japanese #NLP #MeCab
Next was a nice group of short talks on various startups/projects at #Indaba2023. I particularly liked the talks by Asmelash Teka Hadgu and Paul Azunre with targeted #NLP approaches https://www.youtube.com/watch?v=F21GuEZ8EoY (6/11) #startups #Africa
Addendum 1
Instruction tuning: https://en.wikipedia.org/wiki/Large_language_model#Instruction_tuning
https://mastodon.social/@persagen/110945422507756632
* self-instruct approaches
* enable LLM to bootstrap correct responses
FactLLaMA: Optimizing Instruction-Following Language Models with External Knowledge for Automated Fact-Checking
https://arxiv.org/abs/2309.00240
LLaMA: https://en.wikipedia.org/wiki/LLaMA
* family of large language models (LLM) released 2023-02 by Meta AI
#LLM #LLaMA #FactLLaMA #AugmentedLLM #SelfSupervisedLLM #LargeLanguageModels #QuestionAnswering #NLP #GPT
Cade Metz’s NYTimes article on Doug Lenat, featuring quotes from Ken Forbus and @garymarcus. The connection between the game Traveller and #Cyc is interesting and is new to me.
Douglas Lenat, Who Tried to Make Computers More Human, Dies at 72 https://www.nytimes.com/2023/09/04/technology/douglas-lenat-dead.html
#AI #Reasoning #Commonsense #Knowledge #KnowledgeGraph #NLP #NLProc
It's fascinating to see how natural language processing and AI have evolved.
Especially when it comes to evaluating language model responses.
What stands out is the shift from solely relying on quantitative metrics to incorporating more human judgment and nuanced evaluation methods.
In the early days, metrics like BLEU and ROUGE were the go-to tools for assessing language model outputs. While these metrics provide valuable insights, they often couldn't capture the complete picture of response quality.
Now, there's a growing recognition that human evaluators bring an essential cog in the evaluation process. Their ability to assess factors like relevance, clarity, and coherence provides a more holistic view of a model's performance. This shift towards human evaluation reflects a deeper understanding of the nuances of language and the importance of context. (Any linguist will tell you that most computer scientists kinda get language wrong.)
People often talk at cross-purposes without even realizing.
What happens when we do it at scale?
When it comes to NLP, sorting the science from the pseudo-science is the first step toward a meaningful conversation about the ethical concerns we can no longer ignore.
#NLP #programming #linguistics #language #Medium #history #communication
I think this is really clever! The authors of this repo "train" an individual compressor for each of a set of topics of interest. Then, given an input text, they compare each compressor on the input. The topic of the compressor that has the best compression ratio is most likely the topic of the input text. The collection of compressors is the classification model.
11/end
[3] Previously: “Model Editing: Performing Digital Brain Surgery”. https://www.linkedin.com/posts/benjaminhan_llms-causal-papers-activity-7101756262576525313-bIge
[4] Shirui Pan, Linhao Luo, Yufei Wang, Chen Chen, Jiapu Wang, and Xindong Wu. 2023. Unifying Large Language Models and Knowledge Graphs: A Roadmap. http://arxiv.org/abs/2306.08302
10/ REFERENCES
[1] Fabio Petroni, Tim Rocktäschel, Patrick Lewis, Anton Bakhtin, Yuxiang Wu, Alexander H. Miller, and Sebastian Riedel. 2019. Language Models as Knowledge Bases? http://arxiv.org/abs/1909.01066
[2] Vishwas Mruthyunjaya, Pouya Pezeshkpour, Estevam Hruschka, and Nikita Bhutani. 2023. Rethinking Language Models as Symbolic Knowledge Graphs. http://arxiv.org/abs/2308.13676
9/ But LLMs can certainly help in bringing in domain-specific or commonsense knowledge in a data-driven way. In conclusion: why not both [4]? :-)
8/ There are surely other benefits of using KGs to collect and organize knowledge. They do not require costly retraining to update, therefore can be updated more frequently to remove obsolete or incorrect facts. They allow more trackable reasoning and can offer better explanations. They make fact editing more straightforward and accountable (think of GDPR) compared to model editing [3].
7/ Their result shows that even #GPT4 achieves only 23.7% hit@1 on average, even when it scores up to 50% precision@1 using the earlier proposed LAMA benchmark (screenshot). Interestingly, smaller models like BERT can outperform GPT4 on bidirectional, compositional, and ambiguity benchmarks, indicating bigger is not necessarily better.
6/ In each benchmark, instead of asking LLMs to retrieve masked words from a cloze statement, it also asks the LLMs to retrieve all of the implied facts and compute scores accordingly (screenshot).
5/ Instead, they focus on more intricate topological and semantic attributes of facts, and propose 9 benchmarks testing modern LLMs’ capability in retrieving facts with the following attributes: symmetry, asymmetry, hierarchy, bidirectionality, compositionality, paths, entity-centricity, bias and ambiguity (screenshots).
4/ But is that all? A recent paper revisits this question and offers a different take [2]. The authors believe just testing isolated fact retrieval is not sufficient to demonstrate the power of KGs.
3/ The result shows that even without specialized training, language models such as BERT-large can already retrieve decent amount of facts from their weights (screenshot).
2/ An early paper in 2019 [1] posited that compared to #KnowledgeGraphs, it is easier for language models to adapt to new data without human supervision, and they allow users to query about an open class of relations without much restriction. To measure the knowledge encoding capability, the authors construct the LAMA (Language Model Analysis) probe where facts are turned into cloze statements and language models are asked to predict the masked words (screenshot).
1/ In this age of LLMs and generative AI, do we still need knowledge graphs (KGs) as a way to collect and organize domain and world knowledge, or should we just switch to language models and rely on their abilities to absorb knowledge from massive training datasets?
Dog Lenat, founder of #Cyc, passed away earlier this week. From Professor Ken Forbus:
"People in AI often don't give the Cyc project the respect it deserves. Whether or not you agree with an approach, understanding what has happened in different lines of work is important. The Cyc project was the first demonstration that symbolic representations and reasoning could scale to capture significant portions of commonsense…”
https://www.linkedin.com/posts/forbus_ai-knowledgegraphs-krr-activity-7103445990954700800-qcd-
Nach sechs Monaten Arbeit freuen @LStina und ich uns, unser Projekt #Biaslyze, zur Erkennung von #Bias in #NLP Modellen, bei den #PrototypeDemos des @PrototypeFund vorzustellen! :)
Ihr findet mehr Details und Tutorials auf unserer Website https://biaslyze.org.
So I’ve been lurking here some time without a proper #introduction! I’m a linguist working in #NLP but interested in a variety of topics from historical linguistics to language learning and, further afield, SFF, writing, cats, birds, wine, gaming, and any combination thereof 😀
…okay maybe not giving wine to cats BUT STILL
Elon is planning to effectively kill social cards on #Twitter. Social cards were a big part of my dissertation work. I published a few papers about generating them via #ComputerVision, #NLP, and #MachineLearning because they make for nice bits of document #Summarization and #Storytelling. Now Musk wants them gone to force journalists to write articles directly on Twitter.
Ref (paywall): https://fortune.com/2023/08/21/elon-musk-plans-remove-headlines-news-articles-link-shared-on-x-twitter/
Ref (article about paywalled article): https://9to5mac.com/2023/08/21/twitter-to-hide-news-headlines/
"First they came and told us we couldn't speak our language, then they whacked us for speaking our language, now they've taken our language and want us to sell it back to us."
This fantastic kōrero between @keoni and @parismarx on @techwontsaveus is wide ranging and unmissable:
https://techwontsave.us/episode/177_big_tech_wont_revitalize_indigenous_languages_w_keoni_mahelona
#NLP #TeReo #ASR #LanguageRevitalization #LanguageTechnology #Interoperability and more!
My social media research has been published by Elsevier in the journal *Computers and Geosciences*! This is my first published journal article, so I'm pretty excited about it.
It's entitled "Real-time social media sentiment analysis for rapid impact assessment of floods".
It's also gold open access, so you can read it here:
https://doi.org/10.1016/j.cageo.2023.105405
#igotpublished #computerscience #phd #papers #science #research #flooding #ai #nlp
Hey there!
I'm a Computer Science #PhD Student Researcher at the University of Hull.
Project: "Using #BigData and #AI to Dynamically Predict flood risk"
Research Interests: #ai, #nlp, #computervision and multidisciplinary research to save the planet
Interests:
• #computerscience
• #opensource
• Playing #piano @ ~grade 4/5 ish
• #parkour
• Video game + music / art
• Anything #SliceOfLife
• #cats
• More I can't remember right now
she/her | #ActuallyAutistic
My mid-month newsletter has a rundown of the MyHouse.wad Doom transmedia horror saga, with links to Piranesi; scipy tutorial links, some good NLP tool developments, other AI art experiments and updates. #games #aiart #weird #DataScience #nlp https://open.substack.com/pub/arnicas/p/titaa-445-myhouse-is-a-very-weird
What the actual f.
“Low-Resource” Text Classification: A Parameter-Free Classification Method with Compressors
Jiang et al 2023
https://aclanthology.org/2023.findings-acl.426/
"a combination of a simple compressor like gzip with a k-nearest-neighbor classifier ... outperforms BERT on all five OOD datasets, including four low-resource languages. Our method also excels in the few-shot setting, where labeled data are too scarce to train DNNs effectively."
Researchers used artificial intelligence to translate ancient cuneiform texts from Mesopotamia into English. Most of the hundreds of thousands of the cuneiform-inscribed clay tablets remain untranslated. This new neural machine translation model could help us understand more of these ancient texts.
https://www.archaeology.org/news/11449-230518-artificial-intelligence-cuneiform
#AI #ArtificialIntelligence #NLP #Ancient #Language #Histodon
📣 Check out our new paper "DARE: Towards Robust Text Explanations in Biomedical and Healthcare Applications", oral at @aclmeeting by lead author Adam Ivankay this Wednesday!
We show adversarial attacks to #explainability methods for #DeepNeuralNetworks in technical text domains, propose a quantification of this problem, and initial solutions.
📊 Presentation: https://virtual2023.aclweb.org/paper_P1265.html
📄 Paper: https://arxiv.org/abs/2307.02094
💻 Code: https://github.com/ibm/domain-adaptive-attribution-robustness
On My Disk is a personal cloud storage solution that keeps your files secure and private. We are integrating PeARS search into the service to make your files incredibly easy to find, regardless of their physical location.
One ideal shared by PeARS and On My Disk is energy efficiency. So we are dedicated to create 'small AI' solutions that combine old and new machine learning techniques, to create the most climate-friendly NLP tools 🍃 🍐 #smallai #NLP
Hey #NLP—we all know better than to use NLP tools that we can't assess, right? These folks are advertising on the #acl2023nlp Rocket.chat, but give 0 info about how their system was built or evaluated.
The link for http://generalizable.xyz goes to a page that literally just consists of their logo.
In what sense is this GPT? What training data did they use? What test data? What evaluation metric?
But more to the pt: We're here to learn from each other, not to read synthetic versions of papers.
I have quite a lot more followers than I did when I first wrote my introduction, so it’s only fair that I’m writing a new one, bump up the major version.
#introduction #intro #introductions
Hi! o/
I am Ștefan (ș as sh, I also accept Stephan or the equivalent in your language). I’m 21 years old, ♑, he/him, proud #leftist and soon to graduate CS @ UVABc. Sort of proudly living in #romania. My native language is Romanian, fairly proficient at English, slowly learning #finnish (and #italian).
Tried a lot of programming languages in my childhood up until now, a non-chronological list of ones that stuck with me for one reason or another being: VB6 (that’s what I started on at 8 years old), #pascal (+ #freepascal and #delphi), #perl (+ #raku), #tcl #tcltk, #lisp (usually #scheme, on a good day #elisp #emacslisp and #commonlisp), #elixir, #php, #forth, #lua, #oberon, #modula-2, #cpp #c++, #ocaml, #fsharp, #smalltalk (+ #squeak #pharo #self), #ada, #powershell, #dart, #matlab, #rlang, #zig, #nim, #cobol and #julia. I don’t claim full proficiency in all of these, but I’m familiar enough with these (+ some others not mentioned here) that I could get along just fine with 2 weeks at most of studying and looking through cookbooks and examples). I’m flexible in learning new languages and technologies if needed.
I also do #sudoku and #math for fun (especially functional equations and number theory problems, sometimes calculus and geometric algebra). I am interested in #linguists, #conlangs (#lojban and #esperanto) and #nlp, contemporary (post-‘45, usually post-‘89 for me) history, #balkan history, lower-level stuff (I like to learn about how tools around me work, I’m most interested in #compilers, #emulators and #microcontrollers), #typography and #latex, #linux + #bsd, #msdos, #amiga, #oberon, #plan9, #philosophy, #astronomy (especially in a worldbuilding context) and #philosophy, along with other less notable interests.
I engage in #politics relatively often irl, although I’m not inserting it in absolutely every scenario in my life. As I mentioned, I’m a #leftist and #progressive (or… um… #woke) (Nazis and fascists can have a merry fuck off, DNI with me). I am also a spiritual person, a #deist (if you really want to put it that way, an agnostic, although it’s not quite true) and I find #astrology and #tarot interesting (I’ll let you guess my moon and ascendant, let’s see how close you are).
With that being said, I hope I’m welcome here, you can pick your subset of things that interest you from this list, you have plenty of options. :P Quite a bit longer than last time, but oh well…
The {huggingfaceR} #rstats 📦 aims “to bring state-of-the-art NLP models to R. huggingfaceR is built on top of Hugging Face’s transformers library; and has support for navigating the Hugging Face Hub.” By Alex Farach, data scientist at Microsoft, & others. Requires Python & {reticulate}.
https://github.com/farach/huggingfaceR
#NLP @rstats
Oui bonjour c'est enfin l' #introduction (ça fait juste des mois que c'est dans les rappels de mon téléphone)
Ici ça sera sûrement beaucoup #concert , #séries et #cuisine | #pâtisserie
#lecture aussi et sûrement un peu de #train #montagne #homereno (kinda?)
Peut être #rando, #eurovision, #j2s et #jdr 🤷
Et je sais pas si ça transparaîtra vraiment mais team #datascience #nlp tout ça, aussi