Public Alert No. 09/2026.
Public Reminder of NAFDAC’s Regulatory Directive on the Discontinued Registration of Multi-Dose Artemether/Lumefantrine Dry Powder for Oral Suspension
#NAFDACAlerts
https://t.co/wU5rBNmNjs
Yorubas have many gems we don't even talk about.
The late Professor Monsur Akangbe Kenku is finally receiving his due from the new generation of mathematicians from around the globe.
If science class explained everything with Midwest emo music, we would have a lot more people paying attention.
Nuclear power explained.
It’s just a fancy way to heat water and make electricity.
🔊
This is my parting gift for the year.
Arguably my best podcast listen all year. @Nairametrics’ Drinks & Mics wrapped the year with Mayokun Ajibade, SA to the CBN Gov on Financial Markets & Economic Policy.
His clarity reinforces why to bet on PBAT & 🇳🇬.
I strongly recommend.
This guy has finished work here…
Those interest and willing take your pen and paper, take notes and act.
Check him out on TikTok…
I will set you up with information, we are not leaving anyone behind…
Needed to add company knowledge to LLM.
Plan:
- Collect 5,000 company documents
- Convert to training format
- Fine-tune Llama 2 on SageMaker
- Deploy custom model
Started fine-tuning:
- Training time: 6 hours
- Cost: $450 for GPU instances
- Result: Model that knew company facts
But:
- Model hallucinated variations of facts
- Couldn't update without retraining
- New document? Retrain entire model
- Wrong information learned? Retrain entire model
- Each iteration: 6 hours + $450
Then tried RAG (Retrieval Augmented Generation):
- Embedded all documents with OpenAI
- Stored in pgvector (Postgres extension)
- Query flow:
- User asks question
- Find relevant documents (vector similarity search)
- Send documents + question to LLM
- LLM answers using provided context
RAG setup time: 2 hours
RAG cost: $0.02 per query (embedding + LLM)
RAG benefits:
- Update knowledge: Add/remove documents (seconds)
- Fix wrong info: Update document (seconds)
- No retraining needed
- Cite sources (know where answer came from)
- Works with any LLM
Fine-tuning benefits:
- Lower inference cost (no retrieval step)
- Faster responses
- Custom behavior/tone
- Works offline
When I'd use fine-tuning:
- Teaching model new task/format
- Changing model behavior/style
- High-volume inference (cost matters)
- Need offline deployment
When I'd use RAG:
- Adding knowledge that changes
- Need source citations
- Multiple knowledge domains
- Fast iteration needed
Start with RAG, not fine-tuning. Fine-tuning is for behavior, RAG is for knowledge.
You probably don't need custom model weights. You need a better prompt with the right context.
ICAN Scholarships
I wrote about how I completed my ICAN exams on scholarship in the tweet below. The scholarship now covers exam fee and study packs (study/tutorial tuition used to be covered but I understood this is no longer included).
Information on how to access the scholarship can be found here.
https://t.co/7CP1sTKfoR
Three level of scholarships
1. Diamond: First class graduates of any course
2. Gold: Best graduating accounting students
3. Silver: Best 100 level students ( to the ICAN ATS exam)
Applicants must be from an ICAN accredited institutions.
You will generally need the following in addition to other conditions/criteria specified in the link above to apply.
1.Application Letter (indicating your Reg. No., Mobile, contact/postal address and E-mail address)
2. Certificate or Notification of Results (NOT MORE THAN TWO YEARS AFTER SENATE APPROVAL).
3.ORIGINAL Introduction Letter from Your School. Note that the school must be an ICAN-accredited institution as at when the application is submitted.
4. ORIGINAL Recommendation Letter from any of the Institute's District Society .
There are ICAN contacts in the link as well should you need further information.
All the best!
Andrew Ng just dropped a new course on Agentic AI.
Learn to build Agentic Systems that plan, execute, and iterate through multi-step workflows along with tool calls, multi-agent systems and agent evals.
And it's 100% free.