C.O. Becky 柯林斯 贝基 أُعيد تغريده
Gunpowder used to be for entertainment, then it was used to kill.
Drones used to be for entertainment, then they were used to kill.
AI is used for entertainment.
English
C.O. Becky 柯林斯 贝基
371 posts
C.O. Becky 柯林斯 贝基
@CollinceBecky
AI / ML Engineer (2021) CTO & Co-founder, Brance Tech | computer vision startup
Mombasa,Kenya انضم Kasım 2018
96 يتبع77 المتابعون
@Kirembu @akinyi__wendy This is nice work pal. Make it open ill play around with it
English
@akinyi__wendy @CollinceBecky Initially used Grok, then switched to Claude 4.6 to verify and fix what I missed. Here is the corrected Colab notebook. Let me know if it makes sense or if there's something I've missed. I'm not an ML Expert so feedback is welcome!
colab.research.google.com/drive/1VhFmtjv…
English
YouTube
English
@CollinceBecky @Kirembu Keep asking AI . Im glad you are building on it .
English
@Kirembu @akinyi__wendy With better understanding would help
English
@Kirembu @akinyi__wendy Kernels are carefully initialised to convolve over the channels and pick specific image defining features . So some care might need to be taken on initialising the spiral kernel. Ididnt understand the novelness of the residual connection "reverse residual" in this case but ..
English
@TiskTusk Get it from me 1 convolution Layer will never result to anything substantial however cool the math beneath it is. Have worked with mobilenetV2 by itself its abt 50 conv layers with few FC layers.
English
This paper is absolute comedy once you stare at it for a bit. She mentions the architecture having fewer parameters than a Resnet and then proceeds to mention that it is based on a Mobilenet model(so there shouldn't be any bragging about efficiency here, you use mobilenets for constrained compute). And then move on to this section with big claims and see this joke of an "architecture design" section. You can probably brute force your way through it though using intelligent guesses.
Samaritan@0_samaritan
1/ n Hi. I am assuming you are relatively my age. I have read quite a number of research papers on ML and DL . Am also quite familiar with how research works and what it takes to invent a new theory or architecture in ML. Given the things you have listed to have accomplished
English
Mmm ,I beg to disagree. A 0.1 MB FP32 model is only ~25K parameters, which is far too small to produce meaningful DL performance.
For anything resembling an LLM, that scale is essentially non-viable for real language understanding or generation. Claims of performance comparable to 1M+ parameter models would need very strong, reproducible benchmarks to be credible.I would be interested to see the available evidence if there is any.I also stand to be corrected.
English
Oh , and don’t tell me to look for “older” generation . I graduated from Machakos University in 2024 .
I have built Spairally , from conception, to inventing an entirely new high energy physics theory , to inventing an entirely new equation from the theory , to building entirely new kernels , to entirely new deep learning architectures that are less than 0.1MB fp32 - published AI research papers . Trained the models , tested the models , built a very complex backend architecture, built the consumer product .
I know the talent I am looking for . Software engineering is the least of it . At least for now .
English
@vincenteliezer @Qwarqo We also worked on one in 2022. Traction good .Calls from people who wanted houses okey. Popularity pleasing. Now when it came to having reliable inventory i.e The house you see on the site is actually vacant. A hard nut to crack. A unit that is vacant today might not be tommorow
English
every dev has had this 'house hunting app' idea at least once ..
the problem haijawahi kuwa software… kenyans are the edge cases, It’s a spiritual warfare + trust issue stack 🤣
talking from experience - @Qwarqo
KENYA GOSSIP CLUB@kenyagossips
Kenyan X User Urges Developers To Create A House‑Hunting App To Make Relocation Easier For Tenants
English
@weehmzeee Haha we tried in 2022 .The trafic was there .The problem was how to update the inventory . A house thats vacant today wont be tomorrow !
English
Hiii Kenya mlisema hakuna mtu ataunda app ya house hunting
Indonesia
How can they be so smart and damn at the same time . Similar result from chatgpt !
Yoshik@AskYoshik
claude opus 4.7, ladies and gentlemen
English
Unfortunately your time is up. Enda ukalie mayai ocha
ℝ𝕒𝕡𝕦𝕕𝕠🦎@RapudoBrian
Kiongosi kindly support graphics designers 😂😂sii kila time AI
English
@mbiti_mwondi Haha seriously. Its the hardest time to learn programming. Nevertheless he is doing it the right way if he wants to be a competent S.E
English
Ndugu learn how to use AI wachana na hii maneno,time wasting!
Gachie_patrick 🇰🇪🇨🇦@patrickgachie2
Started learning coding under UNIT;STRUCTURED PROGRAMMING mniambie the easiest way to learn this shit
English
@akinyi__wendy @kariuki_shark Lets say its propriatory but 3k . Will shock the best researchers i know
English
Everyone is going big . GPUs , trillion parameters - but at the end of the day , billions of devices are mobile phones , microcontrollers etc with limited memory capacity.
The goal of my research is to build models that run on these devices .
~ I said that today 🌝
English
@akinyi__wendy @kariuki_shark @akinyi__wendy what task are you using it for and whats the role of the model on the task !?
English
@CollinceBecky @kariuki_shark performs better than an 11M parameter one
English
@akinyi__wendy @kariuki_shark Wow 3k ! , im genuinly curious on what meaningful task a model could achieve at that size !
English
@kariuki_shark The models I create are at 3k params … latest one is at 17k params . 500M params is tooooooooo much
English
What happened to Eureka Labs Andrej !!
Andrej Karpathy@karpathy
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
English
Forget gnome , the future of OS UI will be Interactive Chat Interfaces ! - I.C.I
English