Software engineer building AI infrastructure, distributed systems & context systems. Building @Cherragon โ https://xitter.cf/t.co/S0BCwsGBMB
London,Uk
Joined June 2016
- Tweets12.8K
- Following532
- Followers336
- Likes9.4K
Kibbโ๐พโ๐พโ๐พ retweeted
Not to be a fear monger but are we in the end game now?๐
All these tech layoffs are ๐ฅฒ
Kibbโ๐พโ๐พโ๐พ retweeted
Today we're coming out of stealth, introducing Neywa Labs ๐
We built the infrastructure to build and deploy AI that runs on the devices people already own. Private, offline, unlimited.
One mission: make AI available to everyone in the world.
neywalabs.ai/
Kibbโ๐พโ๐พโ๐พ retweeted
Former Special Agent reveals the 3 types of people to watch out for โผ๏ธโผ๏ธ
Kibbโ๐พโ๐พโ๐พ retweeted
OS wey i order from ikorudu just text me โIโm comingโ instead of โAm comingโ
FBI nice try
Kibbโ๐พโ๐พโ๐พ retweeted
Well it's official, BeautifulRust has been the fastest growing open source project I've ever created.
Sometimes you just put something out into the world and find people seem to dig it.
Glad to see so many people love Rust as much as I do, and care about writing clean Rust code.
I think we are moving into a time where poorly written code matters even more because if an agent misunderstands something, you'll burn a lot of tokens doing the wrong thing.
One of the biggest feature requests I've had over the last 24-hours is adding the ability to turn C, C++ and Python code into nice, clean, idiomatic Rust.
So I'm building that out now.
Rust on ๐ค ๐ฆ
github.com/morganlinton/Beauโฆ
Kibbโ๐พโ๐พโ๐พ retweeted
Just found out our across the road neighbors are selling their 100 acre farm complete with remodeled 3 years ago house to their grand nephew for what they paid for it in 1972. He's getting a million dollar lot for $92,000. He's been harvesting their soybeans and corn for the last few years and I'm so excited to see what he does with the place, this is how you set the next generation up for success. They're also gifting him all the equipment, tractors, combine, boom truck, skid steer and who knows what else, their equipment barn is huge.
Kibbโ๐พโ๐พโ๐พ retweeted
4 months ago I cloned Airtable for a client in 5 days because they refused to pay Airtable $500/mo
months later, Iโm still maintaining it
every two weeks itโs another edge case, a UI tweak, or a "quick 10-minute request." Theyโve easily spent 10x more in headaches than the $2,000 they wouldโve paid Airtable
just proving that SaaS isn't dead. Cloning isn't enough, if you want it to be same quality as original it becomes a full time job
and at this point isn't it easier to just pay?
Kibbโ๐พโ๐พโ๐พ retweeted
Age slowly reveals certain truths
- Alcohol loses its appeal
- The gym becomes therapy
- Good sleep feels priceless
- Strong health brings freedom
- Quiet evenings feel better than loud nights
- Fewer friends feel more valuable
- Peace becomes a priority
- Simplicity feels like success
Kibbโ๐พโ๐พโ๐พ retweeted
You see where the narrative โa womanโs money is hers but that of her manโs belongs to both of themโ has gotten to? Why is she entitled to her husbandโs money? Why canโt she use her own resources to help her own family? If the table is turned would she agree? Such AUDACITY!!
A man's wife is always upset with him because he sponsors his parents financially and recently even sponsored some of his siblings through school but he never does same for her family.
She recently brought up that one of her siblings had finished high school and needed help going to the University.
Her husband told her he won't be able to sponsor. Her family should find a way. She felt really hurt about it.
Kibbโ๐พโ๐พโ๐พ retweeted
Yup & tbh Iโve always looked at those spaces as spaces to celebrate black achievements/voices & at best network horizontally not really upwards; granted theyโre not always advertised as that, but pls weโre not kids use discernment & move accordingly.
Kibbโ๐พโ๐พโ๐พ retweeted
Last week, I ran a simple, stupid experiment where I converted a deterministic workflow and trained a 3-billion-parameter model, Qwen-2.5, to do the same.
The workflow was incident triage: parse a noisy alert, check metrics, check recent deploys, read logs, conclude a root cause and runbook, then draft a status update.
Here are the results...
- fine-tuned 3B model hit 93.0% e2e success.
- Cost went down 8x, from $1.85 to $0.22.
- Input tokens dropped from about 5,200 to 450 per incident (-91.3%), ofc.
Also, here's the gist of what I did: removed the 3,500-token SOP from the prompt entirely. The system prompt is now just `incident_triage_v1`, and the procedure lives in the weights.
The key detail was loss masking. Alerts and tool outputs are masked, so the model learns only the assistant decisions and not random telemetry numbers.
I know this is a stupid experiment because I will have to retrain the model if the workflow changes even a bit, but it was genuinely a fun one :)
The repo is called `workflow-slm` and is available on my GitHub. Here is a detailed write-up. Give it a read :)
