Sign in

Data Science for Social Impact

@dsfsi.bsky.social
40 followers 104 following 87 posts

Data Science for Social Impact research group in the CS Department, U of Pretoria. Visit us at www.dsfsi.co.za Socials linktr.ee/dsfsi

PostsRepliesMedia
Data Science for Social Impact @dsfsi.bsky.social · 14h
Read Part 1: www.dsfsi.co.za/blog/afirkal... Speak an African language? Register it: app.afirkallm.org/apply #AfirkaLLM #AfricanNLP #NLP 6/
dsfsi.co.za
AfirkaLLM Part 1: Africa is 0.057% of the web. Here is how we grew it.
Before you can teach a model a language, you need its text. Most of Africa’s data was mislabelled, uncrawled, or never online.
000
Data Science for Social Impact @dsfsi.bsky.social · 14h
The full harvest: 16.5M documents, 18.6B tokens, 256 languages, every licence tracked. Still under 1% of what general-purpose models train on, and Afrikaans and Kiswahili are about half of it. That skew is where Parts 2–4 pick up, over the next few days. 5/
100
Data Science for Social Impact @dsfsi.bsky.social · 14h
Our AfriCC pipeline filtered 892,000 candidate pages to 109,062 confirmed African-language pages, each agreed on by two independent detectors (AfroLID and GlotLID). 4/
100
Data Science for Social Impact @dsfsi.bsky.social · 14h
So we started upstream. 85 contributors across 19 countries curated 700 seed URLs in 34 languages. Common Crawl's late-2025 crawl then saw a 38-fold expansion in African domains. Some languages leapt: Setswana +279%, Sango +259%, Luganda +160%. 🚀 3/
100
Data Science for Social Impact @dsfsi.bsky.social · 14h
The first surprise: much of the "missing" data was never missing. It was mislabelled. A Hausa page tagged as Arabic, or a Chichewa post tagged as English, never gets picked up. For AI purposes it doesn't exist. 2/
100
Data Science for Social Impact @dsfsi.bsky.social · 14h
Africa has 2,000+ languages, yet fewer than 6 in every 10,000 pages in Common Crawl, the web archive most AI learns from, are in an African language. 🌍 Part 1 of our new AfirkaLLM series: where the data comes from. 🧵 @vukosi.bsky.social 1/
211
Data Science for Social Impact @dsfsi.bsky.social · 03/10/2026
"Africa has no shortage of AI talent, groundbreaking research or innovative ideas. So why do so many promising AI innovations never make it into the hands of the people who need them?" Dog-fooding but for African tech with Vukosi Marivate | On What Terms?
youtube.com
Dog-fooding but for African tech with Vukosi Marivate | On What Terms? | Episode 2
Africa has no shortage of AI talent, groundbreaking research or innovative ideas. So why do so many promising AI innovations never make it into the hands of the people who need them? In this…
000
Reposted by Data Science for Social Impact
Vukosi Marivate @vukosi.bsky.social · 18/08/2026
The Deep Learning Indaba is, and always will be, its people 🌍💡 @deeplearningindaba.bsky.social Reflecting on #DLI2026 at Pan-Atlantic University in Lagos, Nigeria, I wrote about what makes this community unique and why its travelling nature matters. Read the post 👇 www.vima.co.za/2026/08/18/t... 1/
vima.co.za
The Deep Learning Indaba is, and always will be, its people [post Deep Learning Indaba 2026]
During the Mutapa Empire, which encompassed parts of what are now Zimbabwe and Mozambique, the royal court, the dzimbahwe, moved from place to place rather than remaining in a single location. As i…
131
Data Science for Social Impact @dsfsi.bsky.social · 25/08/2026
Webinar Invitation - Sunflower 2.0 -- Scaling Multilingual speech and text LLMs to 67 African languages, Dr Ernest Mwebaze, SunbirdAI Date: 3 September 2026 Time: 15:00-16:00 PM SAST RSVP - forms.gle/G6m3zsLUqxWK...
000
Data Science for Social Impact @dsfsi.bsky.social · 01/08/2026
Flashback to our last seminar | 🤖 Is AI Shaping Our Future—Or Are We Shaping AI?
youtube.com
🤖 Is AI Shaping Our Future—Or Are We Shaping AI?
As Artificial Intelligence rapidly moves from theory to reality, we face a critical question: How do we ensure these systems serve the public interest while managing the immense uncertainty of their…
000
Data Science for Social Impact @dsfsi.bsky.social · 01/08/2026
AI is moving fast | #AI4Good Panel with Yoshua Bengio, Vukosi Marivate, Joëlle Barral & Yutaka Matsuo
youtube.com
AI is moving fast | Yoshua Bengio, Vukosi Marivate, Joëlle Barral & Yutaka Matsuo
Four of the world’s foremost AI scientists. One moderator from the generation that will inherit their decisions. Yoshua Bengio, Vukosi Marivate, Joëlle Barral and Yutaka Matsuo, members of the UN…
000
Data Science for Social Impact @dsfsi.bsky.social · 31/07/2026
6/6 🤝 If you are attending #DLI2026 in Lagos, make sure to catch our talks, join our workshops, and visit our posters! Read the full lab announcement on our blog 👇 www.dsfsi.co.za/blog/dsfsi-a...
000
Data Science for Social Impact @dsfsi.bsky.social · 31/07/2026
5/ ⚖️ Poster Judging: PhD Candidate Moyahabo Rabothata will be serving as a Poster Judge at #DLI2026, supporting and evaluating emerging researchers from across the continent.
100
Data Science for Social Impact @dsfsi.bsky.social · 31/07/2026
4/ 📊 Poster Showcase: Drop by the poster sessions to chat with our researchers! • Dr Seani Rananga: Detecting misinformation in African languages using LLMs. • Nontokozo Manukuza: Interpreting isiZulu idiomatic expressions using LLMs.
100
Data Science for Social Impact @dsfsi.bsky.social · 31/07/2026
3/ 🛠️ Workshops & Tutorial: Postdoc Dr Idris Abdulmumin is co-organising two workshops on community-led dataset collection & the AfricaNLP Playbook. He will also deliver a hands-on tutorial on OCR and Audio Transcription challenges for real-world African data! 💻
110
Data Science for Social Impact @dsfsi.bsky.social · 31/07/2026
2/ Plenary Keynote & Accepted Paper: Prof. @vukosi will deliver a main plenary keynote on "What Do Our Benchmarks Actually Measure? Evaluation Challenges for African Language AI" (Wed, 5 Aug). Also has paper "The Annotation Scarcity Paradox in Low-Resource NLP Evaluation" 📄
100
Data Science for Social Impact @dsfsi.bsky.social · 31/07/2026
1/🇳🇬 The countdown is on for #DLI2026 in Lagos, Nigeria! The @dsfsi_up lab is heading to the Deep Learning Indaba (2–7 August 2026). Here is a thread of where you can find our team, research, and workshops throughout the week! 👇 www.dsfsi.co.za/blog/dsfsi-a...
231
Reposted by Data Science for Social Impact
Ben Waber @bwaber.bsky.social · 09/06/2026
Next was a compelling talk by Shaimaa Lazem on community-centered practices for African language digitization at @dsfsi.bsky.social www.youtube.com/watch?v=eoLM... (7/8)
youtube.com
Decolonising AI: Community-Centred Practices for African Language Digitisation | Prof. Shaimaa Lazem
YouTube video by Data Science for Social Impact (DSFSI)
111
Reposted by Data Science for Social Impact
Ben Waber @bwaber.bsky.social · 12/06/2026
Next was an interesting talk by Jean Louis Fendji on the datafication blind spot in African AI policy at @dsfsi.bsky.social www.youtube.com/watch?v=kVew... (13/14)
youtube.com
Beyond Dataset Creation:The Datafication Blind Spot in African AI Policy-Jean Louis Fendji
YouTube video by Data Science for Social Impact (DSFSI)
111
Data Science for Social Impact @dsfsi.bsky.social · 27/07/2026
Email pls dsfsi.info@up.ac.za
000
Reposted by Data Science for Social Impact
Vukosi Marivate @vukosi.bsky.social · 18/05/2026
From Morekolodi Primary & Tsogo High, to WitsUniversity & RutgersU, to my current daily homes at U of Pretoria & Lelapa AI —thank you. To the pioneering students, lab members @dsfsi.bsky.social , and community builders at @deeplearningindaba.bsky.social & MasakhaneNLP: you are the actual engine.
111
Reposted by Data Science for Social Impact
Vukosi Marivate @vukosi.bsky.social · 18/05/2026
I don't see a personal achievement. I see a mirror reflecting a vast village. I am deeply mindful of the principle I was raised with: "To whom much is given, much is expected." This honour belongs to the spaces and people that nurtured my mind over the last 40 years.
121
Data Science for Social Impact @dsfsi.bsky.social · 13/05/2026
This work is crucial to ensuring that the future of AI in South Africa is not just technologically advanced, but ethically grounded, legally sound, and socially impactful. 🌍✨ #AI #AIGovernance #SouthAfrica #DigitalTransformation #DataScience (5/5)
000
Data Science for Social Impact @dsfsi.bsky.social · 13/05/2026
This multidisciplinary panel ensures SA's AI policy is evidence-based and strategically aligned. The panel includes: • Prof. B. Rosman (Chair) • Prof. V. Marivate • Prof. A. Gildwald • Ms. H. Irvine • Dr. T. Feela • Dr. J. Mtsweni • Adv. L. Tshikalange (4/5)
100
Data Science for Social Impact @dsfsi.bsky.social · 13/05/2026
In another significant move, Prof. Marivate has been appointed to the DCDT Independent Expert Review Panel to assist in reviewing national AI policy. ⚖️ (3/5)
100
Data Science for Social Impact @dsfsi.bsky.social · 13/05/2026
On 15 May 2026, Prof. Marivate will deliver a public lecture ahead of the DSTI Budget Vote. Title: "Harnessing AI for Inclusive and Sustainable Socioeconomic Development" 💡 A vital conversation on aligning AI with our nation's development goals. (2/5)
100
Data Science for Social Impact @dsfsi.bsky.social · 13/05/2026
Big news for the South African AI landscape! 🇿🇦 🚀 We are excited to announce two major developments involving our PI, Prof. Vukosi Marivate, regarding AI governance and policy in South Africa. 🧵 (1/5) @UPTuks @dstigovza @CommsZA
100
Data Science for Social Impact @dsfsi.bsky.social · 05/05/2026
Watch Matthew Smith's guest talk here: 👉 www.youtube.com/watch?v=wKLv... Check out the full event recap: 👉 www.dsfsi.co.za/ai4d-languag... Join us Thursday for our final retrospective post: Ethics, Law & Deployment! 4/
000
Data Science for Social Impact @dsfsi.bsky.social · 05/05/2026
For the Global South, where language AI is closing the digital divide, integrity is everything. We cannot afford to build our knowledge systems on "AI-confident" misinformation 3/ @uptuks
100
Data Science for Social Impact @dsfsi.bsky.social · 05/05/2026
Hallucinated references aren't just an annoyance, they are a threat to our scientific credibility. Matthew Smith explains how we can develop "automated verification" tools to ensure that the facts and citations provided by AI are anchored in real-world, verified databases 2/
100
Data Science for Social Impact @dsfsi.bsky.social · 05/05/2026
If our AI research tools invent facts, how can we trust the science? 🔎 In the 6th instalment of our 2026 @AI4Dev Workshop retrospective, we’re spotlighting Matthew Smith (IDRC) and his talk on AI-driven research integrity. 🧵 1/
100
Data Science for Social Impact @dsfsi.bsky.social · 30/04/2026
From nuance to resilience—we are building the next generation of African AI infrastructure. Devon - www.youtube.com/watch?v=vlL1... Tsosheletso - www.youtube.com/watch?v=u8Hf... Check out the recap: 👉 www.dsfsi.co.za/blog/ai4d-25... Join us next Thursday as we wrap up with Ethics & Law! 4/4
youtube.com
The Hidden Danger of AI: Understanding "Model Collapse" | AI4D African Languages Workshop Mar 2026
Title: The Hidden Danger of AI: Understanding "Model Collapse" | Dr. Devon Jarvis Description: As Large Language Models (LLMs) continue to ingest vast amounts of synthetic data from the internet, we…
000
Data Science for Social Impact @dsfsi.bsky.social · 30/04/2026
Tsosheletso Chidi (UP) reminded us: "People don't stay without care." She challenged us to build AI that is stewarded, not just extracted. The goal isn't just a perfect model—it's building trust and meaningful partnerships with the communities holding our cultural heritage. 3/
100
Data Science for Social Impact @dsfsi.bsky.social · 30/04/2026
Devon Jarvis (Wits) unpacked "Model Collapse." When LLMs are trained on synthetic data, they lose expressivity, language becomes repetitive, and cultural nuance for low-resource languages is the first thing to vanish. We are building systems that risk "forgetting" their roots 2/
100
Data Science for Social Impact @dsfsi.bsky.social · 30/04/2026
What happens when our models start "eating their own tails"? 📉 In the 5th installment of our 2026 AI4D Workshop retrospective, we’re tackling the "Last Mile"—the hard path from a research paper to a system that actually serves society. 🧵 1/ @UPTuks @AI4Dev
100
Data Science for Social Impact @dsfsi.bsky.social · 23/04/2026
The future? Interdisciplinary labs that merge cybersecurity, law, and language AI. Watch the full spotlight session here: 👉 www.youtube.com/watch?v=0bD9... Check out the full workshop archive: 👉 www.dsfsi.co.za/blog/ai4d-25... (4/4) #InfoSec #AfricanAI #DSFSI #AIResearch
000
Data Science for Social Impact @dsfsi.bsky.social · 23/04/2026
The threat landscape is changing fast: ❌ Data Poisoning ❌ AI-Enabled Scams ❌ Synthetic Media in legal evidence Prof. Venter warns that our current legal frameworks—which were designed for humans—are not yet equipped to hold "black box" AI models accountable. 3/
100
Data Science for Social Impact @dsfsi.bsky.social · 23/04/2026
Forensic investigators in remote areas often hit a "language wall." When evidence is buried in local, culturally nuanced languages, conventional forensic tools fail. We need AI that understands these languages to ensure digital justice is accessible to all. 2/
100
Data Science for Social Impact @dsfsi.bsky.social · 23/04/2026
If the first era of AI was defined by scale, the next must be defined by resilience. 🛡️ In our 4th workshop retrospective, we’re spotlighting Prof. Hein Venter’s talk on the intersection of #Cybersecurity and #NLP. 🧵 1/ @UPTuks
100
Reposted by Data Science for Social Impact
Vukosi Marivate @vukosi.bsky.social · 19/04/2026
‘To Bring In More Voices And Perspectives’: The South African Professor Refusing To Let The Continent Be A Mere Spectator In The AI Race www.forbesafrica.com/current-affa...
1134
Reposted by Data Science for Social Impact
Vukosi Marivate @vukosi.bsky.social · 21/04/2026
In relation to the World Development Report (WDR) 2026, please read Statement from the African AI Ecosystem Session on the sidelines of the WDR 2026 Africa AI Consultation which was facilitated by AfriDSAI www.up.ac.za/afridsai/sta...
up.ac.za
Statement from the African AI Ecosystem Session on the sidelines of the WDR 2026 Africa AI Consultation
AfriDSAI – World Bank WDR 2026 Consultation | March 20, 2026
032
Reposted by Data Science for Social Impact
Vukosi Marivate @vukosi.bsky.social · 22/04/2026
My comments on the South African Draft AI Policy <> SA introduces draft AI regulation policy youtu.be/Ix0nWWQYrMs?...
011
Data Science for Social Impact @dsfsi.bsky.social · 21/04/2026
Innovation is about solving real-world problems. Watch the full session on Applied AI & Systems: 👉 www.youtube.com/watch?v=Baw8... See our full research archive and future roadmap here: 👉 www.dsfsi.co.za/blog/ai4d-25... Join us next Thursday as we spotlight our Cybersecurity in NLP session! 5/
001
Data Science for Social Impact @dsfsi.bsky.social · 21/04/2026
AI can also save lives and increase accessibility🖐️ 🔸 Dinorego Mphogo is building frameworks for multilingual Air Traffic Control to improve aviation safety. 🔸 Thapelo Sindane introduced Dipolelo, a benchmark dataset for South African Sign Language to foster social inclusion 4/
100
Data Science for Social Impact @dsfsi.bsky.social · 21/04/2026
Data is the backbone of financial trust. 📈 Miehleketo Mathebula presented a system to automate ESG data extraction from the messy, unstructured reports of South African organisations. This moves us toward more transparent, data-driven financial intelligence. 3/
100
Data Science for Social Impact @dsfsi.bsky.social · 21/04/2026
Nontokozo Manukuza is tackling the "nuance gap." 🗣️ Large Language Models often struggle with the depth of isiZulu idioms. Nontokozo's work is testing how we can fine-tune models to capture cultural wisdom, not just literal translations. for accurate, culturally-aware response 2/
100
Data Science for Social Impact @dsfsi.bsky.social · 21/04/2026
How do we move AI from the lab to the real world? 🌍 In our third retrospective post from the 2026 AI4D Workshop, we’re spotlighting "Applied AI & Systems." We’re applying NLP to aviation safety, financial reporting, accessibility, and cultural preservation. 🧵 1/ @UPTuks @AI4Dev
100
Data Science for Social Impact @dsfsi.bsky.social · 16/04/2026
The technical gap is closing, one research paper at a time. Watch the full technical rapid talks session here: 👉 www.youtube.com/watch?v=MjwO... Catch up on our research and join our community: 👉 www.dsfsi.co.za/blog/ai4d-25... See you next Tuesday for Part 2: Applied AI & Systems! 5/ @AI4Dev
000
Data Science for Social Impact @dsfsi.bsky.social · 16/04/2026
We also heard from Abebe Tegene and Penelope Matloga: 🔸 Abebe is innovating word embeddings to capture the unique morphology and tone of African languages. 🔸 Penelope is solving the "label noise" problem in code-switched data using noise-aware pseudo-labelling. 4/ @UPTuks
100
Data Science for Social Impact @dsfsi.bsky.social · 16/04/2026
Next: Fiskani Banda on the failures of RAG systems. 🌾 When asking about South African farming, RAG models often hallucinate or latch onto irrelevant keywords. Fiskani is establishing best practices to ensure retrieval mechanisms are actually informative, not just thematic. 3/
100