Amy Da Fucq

[Amy Da Fucq @SprangersAmy]

📊 32 Research Tools to Level Up Your Paper (2026)

From literature review to publication — every tool you need for efficient, productive research.

✅ Free tools
✅ Paid tools (worth it)
✅ All pricing verified

Full guide on Medium 👇

#Research #AcademicTwitter #DataScience

[Amy Da Fucq @SprangersAmy]

👉 https://t.co/QnDb6EfNuP

With complete data ..

[Amy Da Fucq @SprangersAmy]

I wasted ~$100+ testing GPU clouds so you don't have to

47 hours, 5 providers, 1 Llama-3-70B fine-tuning

winner: GPUhub at $16.92
loser: Lambda at ~$45 (storage fees got me)

wrote it all up here:
https://t.co/e2D1WC4FCP

hope it helps 🙏

[Amy Da Fucq @SprangersAmy]

@NaraBuildAI alpha-hawk-9340

[Shaddowles | AI & GPU @Shaddowles]

Most “AI case studies” I see feel like marketing slides.

The ones I actually care about are the simple, honest ones:
– here’s what we tried
– here’s how long it took
– here’s how much VRAM and money it actually used

@hub_gpu is starting to collect stories in that direction — more “real experiments”, less buzzwords. If you’re into that kind of thing, worth keeping an eye on 👇
https://t.co/3xA6QQHnng

#MachineLearning #CloudGPU #MLOps

[Shaddowles | AI & GPU @Shaddowles]

For me, the win isn’t just the hardware, it’s the workflow:
– spin up a GPU
– run a focused experiment
– log time / VRAM / $
– shut it down

I’ve been using GPUHub for this pattern:
https://t.co/McuIIxCYwu

Treat it like a lab bench, not a forever cluster.

[Shaddowles | AI & GPU @Shaddowles]

This is what my “ML lab” looks like now:
– modest machine at home
– rent a 24GB GPU only when I actually need it
– run YOLO/SDXL/LLM experiments end‑to‑end, then shut it down

Instead of a 4090 in my room, I get a GPU I can turn on/off like this 👇

[GPUHub @hub_gpu]

Trained YOLOv8 on VisDrone with an RTX 5090. No instability. No wasted time. Dedicated GPUs hit different. #AI #MachineLearning #YOLOv8 #GPU #DeepLearning

https://t.co/sk5AMXYo01

[GPUHub @hub_gpu]

Can AI understand code from screenshots? Yes. We tried it with Qwen2-VL-2B.
https://t.co/YHnN5omCqw

[Shaddowles | AI & GPU @Shaddowles]

building a large AI model? GPUhub gives you access to RTX
5090/4090/4080 and pro-tier GPUs (RTX Pro 6000 96GB, A800 80GB)
with per-second billing plus daily/weekly/monthly reservations. Every instance includes 50 GB of storage and unlimited free egress, so data prep and dataset transfers stay predictable.

Singapore-based data centers keep latency low for APAC teams, and docs walk you through spinning up your training or inference pipeline in minutes. https://t.co/mppKZkcSar

#AIbuilders #GPUcloud

[Amy Da Fucq @SprangersAmy]

@anjalinirwal02 Hot take: not everyone needs to buy expensive hardware anymore.
If you only need GPU power when you actually use it, cloud GPU makes way more sense.
GPUhub is worth a look.
🔗 https://t.co/2A7GjLHIz7

[Amy Da Fucq @SprangersAmy]

@FurkanGozukara Hot take: not everyone needs to buy expensive hardware anymore.
If you only need GPU power when you actually use it, cloud GPU makes way more sense.
GPUhub is worth a look.
🔗 https://t.co/2A7GjLHIz7

[Amy Da Fucq @SprangersAmy]

@sickdotdev Hot take: not everyone needs to buy expensive hardware anymore.
If you only need GPU power when you actually use it, cloud GPU makes way more sense.
GPUhub is worth a look.
🔗 https://t.co/2A7GjLHIz7

[Amy Da Fucq @SprangersAmy]

@impeculiar1b Hot take: not everyone needs to buy expensive hardware anymore.
If you only need GPU power when you actually use it, cloud GPU makes way more sense.
GPUhub is worth a look.
🔗 https://t.co/2A7GjLHIz7

[Amy Da Fucq @SprangersAmy]

@Pranto39 Hot take: not everyone needs to buy expensive hardware anymore.
If you only need GPU power when you actually use it, cloud GPU makes way more sense.
GPUhub is worth a look.
🔗 https://t.co/2A7GjLHIz7

[Amy Da Fucq @SprangersAmy]

@Sekharendu60107 @X Hot take: not everyone needs to buy expensive hardware anymore.
If you only need GPU power when you actually use it, cloud GPU makes way more sense.
GPUhub is worth a look.
🔗 https://t.co/2A7GjLHIz7

[Shaddowles | AI & GPU @Shaddowles]

I ran a small experiment using a vision-language model to analyze images and charts in real-time.

Instead of writing everything here, I compiled the full setup, results, and cost breakdown into a short PDF.

Key highlights:

- Real-time image + chart analysis
- Stable performance (~500 images in ~30 mins)
- Total cost: ~$1.82

Full breakdown (PDF):
https://t.co/MosPSPKH1U

Would love feedback or thoughts from anyone working with multimodal models.

#AIイラスト #machinelearning #comfyui #ai

[Shaddowles | AI & GPU @Shaddowles]

Stop wasting money buying GPUs.
You can literally rent one in seconds.
👇👇
https://t.co/McuIIxCYwu

[Shaddowles | AI & GPU @Shaddowles]

Hot take:

Most people “doing AI” are just running toy datasets.

I trained YOLOv8 on VisDrone using RTX 5090.

→ 100 epochs in ~1 hour
→ mAP50: 0.41
→ Cost: ~$1.2

This is what real ML looks like.

Thread 👇
#AI #MachineLearning #YOLO