alexis sabourin
252 posts

alexis sabourin
@ctrlsombre
surviving @uwaterloo cs
waterloo, on, ca Katılım Şubat 2026
31 Takip Edilen30 Takipçiler

ai model training isn't just data and gpus
look at the most common training techniques below and see how much they impact the final model behavior:
supervised
"trains a model on labeled input-output pairs so it learns to map new inputs to correct outputs"
- spam detection
- loan default prediction
- sentiment analysis
unsupervised
"finds hidden patterns or structure in unlabeled data without being told what to look for"
- customer segmentation
- anomaly detection
- topic modeling
self-supervised
"generates its own labels from the raw data itself"
- llms
- image representation learning
- contrastive learning
reinforcement
"trains an agent to make sequential decisions by rewarding or penalizing outcomes in an environment"
- game playing
- robotics
- trading algorithms
semi-supervised
"combines a small amount of labeled data to improve learning when labels are expensive to obtain"
- medical imaging with scarce labels
- web content classification
- speech recognition for rare languages
transfer
"takes a model pretrained on one task and fine-tunes or adapts it to a different but related task"
- custom image classifier on a small dataset
- domain-specific language model
- multilingual models
but it doesn't stop there, you can also synergize some of these techniques in order to get specialized and personalized behavior. you do that through a training pipeline with multiple training steps
English

my 𝕏 timeline is filled with people arriving in waterloo for ⁂ socratica symposium 2026 ⁂, that's when you know it's going to be big
you need to make it here asap or you're ngmi, pull up or fomomaxx
symposium.socratica.info
English

@ay_ushr they took the "sell shovels, not gold" advice to heart
English

i'm currently spending 2 hours every day deep working to improve my technical skills and develop my entrepreneurship mindset
- 30 minutes of paul graham essays reading and taking note for later further processing and catching back
- 30 minutes of anthropic readings (academy, engineering, blog, research, build with claude, use cases)
- 1 hour of deepening practical technical skills by doing things, currently working on an ai engineer roadmap and learning how to train ai models from scratch, article coming soon btw
i'm balancing this with university and side commitments, will ramp up when this semester is done
goal is to become cracked
English

@MartinTale how can you even prove that you're not a bot anymore, we really need a solution
English

@vivoplt that's not true, it won't replace you in the near future but at some point when robotics catch up, most of these are gone
English

@Vesper_2005 anyone can take care of themselves when they're "sick". if you are really really sick, you're at the hospital
English

@designbyaron preach, there should be a tool out there to block doom slop
English

@Codie_Sanchez so basically you are already wealthy?
my 2026 wealth gameplan:
- inherit wealth from my parents
English

@KeruboSk common sense? also snoozing doesn't improve your sleep, it just feels kind of good
when you realize that it's an illusion, just stop losing and wasting time
English











