Joined September 2023
- Tweets 518
- Following 622
- Followers 188
- Likes 2,924
62 Photos and videos
Jun 13
this all comes at a cost !
Jun 13
We've reset 5-hour and weekly rate limits for all users.
1
18
Nawin retweeted
Jun 13
The United States is no longer the best place to build an AI lab.
564
387
8,940
449,111
Nawin retweeted
Jun 12
Building has never been this fun! We have been constantly building Dolphinflow.
Mainly listening to users and we have officially found our niche: Traders
Traders have a solid understanding of market but they need workflows that can be replicated. That's what @dolphinflow_ is.
7
3
55
2,217
Jun 11
AI has made it much easier to pick up new languages and get things working quickly, but understanding why something works (or doesn't) is still on us. The syntax barrier is lower than ever, but intuition, debugging skills, and good engineering judgment still needed.
Jun 11
In my Applied AI Cohort, at the end, when we chit-chat, someone asked this question: Given that AI is generating most of the code, is it even useful to learn new languages deeply? Here is my take.
You should learn at least one programming language really well - like, really, really well - and know the specifics of others.
For example, if you are building something in Go for the first time, you can get far by prompting your way through it. You can ship working code without knowing the basics.
But yes, you can read the code, but can you truly understand it? Let me double down...
Go has goroutines and channels. Rust has ownership and lifetimes. These are not syntax details; they are mental models.
AI can generate code in a language you do not know, but it cannot generate intuition for a language you have never understood.
When a deadlock happens, memory usage balloons, or a race condition surfaces, you need that mental model to reason about it. AI cannot hand you that reasoning.
According to me, what AI changes is the ramp-up time - more specifically, syntax friction. That cost is now close to zero. The remaining work is the interesting part: building intuition for how the language thinks.
So, you do not need to memorize every standard library method in every language. But you do need to deeply understand at least one language and then pick up the specifics of others as needed.
That depth in one language gives you the mental model to reason about all the others.
As engineers, our job is to solve problems, not necessarily to write code.
Hope this helps.
1
7
Jun 11
If you don't build your own dream, someone will hire you to build theirs.
~ Jim Rohn
2
16
Nawin retweeted
Jun 9
This is a super exciting release - Claude Fable 5 is the same underlying model as Mythos but with added safeguards. The benchmarks are great and it's SOTA on everything by a margin but I'll add that *qualitatively* also, this is a major-version-bump-deserving step change forward (imo of the same order as Claude 4.5 was in November), peaking especially for long problem-solving sessions on very difficult problems. You can give it a lot more ambitious tasks than what you're used to, the model "gets it" and it will just go, and it's never felt this tempting to stop looking at the code at all (but don't do this in prod!). The model still has quirks that people will run into and the safeguards are configured to be a little too trigger happy for launch, which can hopefully be tuned over time.
I feel a lot of things changing as working software increasingly comes out on a tap. The Jevon's paradox kicks in and I feel my own demand for software growing substantially. You can ask for anything - explainers, visualizers, dashboards, bespoke single-use apps (e.g. a full wandb that is hyper-specific just for your project), you can 10X your test suite, auto-optimize code, run giant research projects with custom HTML for the results, anything! "Free your mind" (Matrix ref). Really looking forward to all the things people build!
Fable 5 is state-of-the-art on nearly all tested benchmarks, with exceptional performance in software engineering, knowledge work, scientific research, and vision.
The longer and more complex the task, the larger Fable 5’s lead over our other models.
ALT Benchmark table titled Mythos 5 & Fable 5, comparing Claude Mythos 5 and Fable 5 against Claude Mythos Preview, Claude Opus 4.8, GPT 5.5, and Gemini 3.1 Pro.
1,269
2,368
25,272
2,686,112
Jun 9
Anthropic just released Claude Fable 5.
At first glance, it looks like another model launch.
It's actually much more interesting than that......................
1
2
57
Jun 9
The bigger story isn't the benchmark scores.
It's the release strategy.
We're seeing AI companies move from:
"Build the smartest model possible"
to
"Build the smartest model that can be deployed responsibly."
1
1
15
Jun 9
Claude Fable 5 may be remembered as one of the first major examples of capability and safety being developed together rather than separately.
And that's probably a preview of how future frontier AI systems will be released.
1
45
Jun 8
Memory safety without a garbage collector feels like one of those engineering ideas that sounds obvious only after someone figures it out....
1
3
42
Jun 8
Before Rust, the trade-off seemed pretty clear: you either got the convenience of a garbage collector or the performance and control of manual memory management.
Rust challenged that assumption.
1
1
37
Jun 8
What I find most interesting isn't just the feature itself, but the fact that so much correctness is enforced before the program even runs.
The more I learn about it, the more it feels like a clever piece of engineering rather than just a language feature. #rust
1
29
Nawin retweeted
Jun 6
the more i work with AI the more i realise how important (and sort of exhausting) it is to continuously steer them towards the right way of engineering
if you're not looking, you end up with 100s of one-time usable functions, weird ass approaches and tests, code that sort of works but looks broken
it hurts your future shipping velocity if you don't get those things right from get go.
Jun 6
I would just like to say that after approximately 15 separate prompts I have been able to achieve chat GPT producing a relatively small amount of code in a fashion that I approve of.
Thank you, thank you very much. The code of course was only 45 lines long, spent an aggregate time of 1 hour on it, could have been done in about 6 minutes.
This. Is. Progress.
20
15
229
16,257