kiwi 🥝

Question everything.
handcash $kiwi251

[ada. @adaezeokaro]

To anyone finding their fire again, may it come back steady, unshakable, and impossible to lose.

[Beka @bekacru]

I'm happy to announce @better_auth has raised a $5M seed led by PeakXV Partners (formerly Sequoia Capital India & SEA), with participation from Y Combinator, Chapter One, P1 Ventures, and a group of incredible investors and angels

We're making rolling your own auth not just doable, but the simplest and obvious choice at any scale

[Rohan Paul @rohanpaul_ai]

⚙️ The Experimental Setup

Fifty-four Boston-area students wrote SAT-style essays under three conditions: ChatGPT only, Google only, or brain only.

Each person completed three timed sessions with the same condition, then an optional fourth session in the opposite condition.

A 32-channel Enobio headset recorded brain signals throughout, and every keystroke, prompt, and interview answer was archived for analysis.

[ₕₐₘₚₜₒₙ @hamptonism]

i stay humble because I’ve seen people become what they looked down on.

[fekede agwar @agwar_fekede]

In a six hours long heart surgery local cardiac team of Ethiopia successfully performed the first Sub-Mitral aneurysm repair along with Mitral valve replacement for a 31 years old patient.

[Shega Media @shegamedia]

Better Auth, a 7-month-old open-source authentication framework for TypeScript, developed by two Ethiopians, has been selected for Y Combinator's Spring 2025 Cohort, securing $500,000 in seed funding.

https://t.co/SJBZUPxcJv

[Andrej Karpathy @karpathy]

When working with LLMs I am used to starting "New Conversation" for each request.

But there is also the polar opposite approach of keeping one giant conversation going forever. The standard approach can still choose to use a Memory tool to write things down in between conversations (e.g. ChatGPT does so), so the "One Thread" approach can be seen as the extreme special case of using memory always and for everything.

The other day I've come across someone saying that their conversation with Grok (which was free to them at the time) has now grown way too long for them to switch to ChatGPT. i.e. it functions like a moat hah.

LLMs are rapidly growing in the allowed maximum context length *in principle*, and it's clear that this might allow the LLM to have a lot more context and knowledge of you, but there are some caveats. Few of the major ones as an example:

- Speed. A giant context window will cost more compute and will be slower.
- Ability. Just because you can feed in all those tokens doesn't mean that they can also be manipulated effectively by the LLM's attention and its in-context-learning mechanism for problem solving (the simplest demonstration is the "needle in the haystack" eval).
- Signal to noise. Too many tokens fighting for attention may *decrease* performance due to being too "distracting", diffusing attention too broadly and decreasing a signal to noise ratio in the features.
- Data; i.e. train - test data mismatch. Most of the training data in the finetuning conversation is likely ~short. Indeed, a large fraction of it in academic datasets is often single-turn (one single question -> answer). One giant conversation forces the LLM into a new data distribution it hasn't seen that much of during training. This is in large part because...
- Data labeling. Keep in mind that LLMs still primarily and quite fundamentally rely on human supervision. A human labeler (or an engineer) can understand a short conversation and write optimal responses or rank them, or inspect whether an LLM judge is getting things right. But things grind to a halt with giant conversations. Who is supposed to write or inspect an alleged "optimal response" for a conversation of a few hundred thousand tokens?

Certainly, it's not clear if an LLM should have a "New Conversation" button at all in the long run. It feels a bit like an internal implementation detail that is surfaced to the user for developer convenience and for the time being. And that the right solution is a very well-implemented memory feature, along the lines of active, agentic context management. Something I haven't really seen at all so far.

Anyway curious to poll if people have tried One Thread and what the word is.

[Juhana Myllys @JuhanaMyllys]

When a 2D game works in 4D.

[kiwi 🥝 @kiwi_251]

I wanted to report an account for impersonation but X did not allow me to search the user being impersonated while filling the report since the user is blacklisted. I hope X can fix this issue by allowing users to search even blacklisted accounts when filling a report.

[Michael Hla @hla_michael]

I taught an LLM to optimize proteins. It proposed a better carbon capture enzyme.

Introducing Pro-1, an 8b param reasoning model trained using GRPO towards a physics based reward function for protein stability.

It takes in a protein sequence + text description + previous experimental results, reasons over the information given in natural language, and proposes modifications to improve the stability of the given sequence.

🧵(1/n)

[the vibe formerly known as jules @echotrx]

These vids of DJs playing YT to mp3 songs made shitter thru Stems compression with reactions of the modern BIPOC Yuppie audience are a COINTELPRO from record labels to dilute dance music and re-popularize older songs to the profit of private equity groups that bought the songs.

[Obsidian @obsdmd]

Obsidian is now free for work.

Starting today, the Obsidian Commercial license is optional. Anyone can use Obsidian for work, for free. If Obsidian benefits your organization, you can still purchase Commercial licenses to support development.

Nothing else is changing. No account required, no ads, no tracking, no strings attached. Your data remains fully in your control, stored locally in plain text Markdown files. All features are available to you for free without limits.

Why make this change? Simplicity. The Commercial license terms were confusing and added unnecessary complexity to our pricing. Furthermore, as the Obsidian Manifesto states: "we believe that everyone should have the tools to think clearly and organize ideas effectively". This change brings us closer to that principle.

People in over 10,000 organizations use Obsidian. Many work in high-security environments, like government, cybersecurity, and finance. Some of the largest organizations in the world, including Amazon and Google, have thousands of employees using Obsidian every day. These teams rely on Obsidian to think more effectively and keep total ownership over private data.

Previously, people at companies with two or more employees were required to purchase a Commercial license to use Obsidian for work. Going forward, the Commercial license is no longer required, but remains an optional way for organizations to support Obsidian, similar to the Catalyst license for individuals.

Organizations that support Obsidian are now featured on the Obsidian Enterprise page. Your organization can be showcased by purchasing 25 licenses or more.

Along with Commercial and Catalyst support, our add-on services, Sync and Publish help Obsidian remain 100% user-supported. In the future, we hope to offer more services designed for teams. As always, these will be optional.

[Thomas Wolf @Thom_Wolf]

After 6+ months in the making and burning over a year of GPU compute time, we're super excited to finally release the "Ultra-Scale Playbook"

Check it out here: https://t.co/dekxY4BQZO

A free, open-source, book to learn everything about 5D parallelism, ZeRO, fast CUDA kernels, how and why overlap compute & communication – all scaling bottlenecks and tools introduced with motivation, theory, interactive plots from our 4000+ scaling experiments and even NotebookLM podcasters to tag along with you.

- How was DeepSeek trained for $5M only?
- Why did Mistral trained an MoE?
- Why is PyTorch native Data Parallelism implementation so complex under the hood?
- What are all the parallelism techniques and why were they invented?
- Should I use ZeRO-3 or Pipeline Parallelism when scaling and what's the story behind both techniques?
- What is this Context Parallelism that Meta used to train Llama 3? Is it different from Sequence Parallelism?
- What is FP8? how does it compares to BF16?

In this book, our goal was to gather, in a single place, a coherent, easy to read yet detailed story of all the techniques that make today's LLM scaling possible.

The largest factor for democratizing AI will always be teaching everyone how to build AI and in particular how to create, train and fine-tune high performance models. In other word making accessible to everybody the techniques that power all recent large language models and efficient training is possibly one of the most essential of them.

What started as a simple blog-post ended up becoming an interactive writing piece containing 30k+ words. So we've decided to actually print it as a real 100-pages physical book as well: the physical ultrafast playbook –containing all the science of distributed and fast AI training.

We plan to send free copies as gifts to the first readers of the online version so feel free to add your email in the form linked in the blog post.

[Babbage | BRC100 @ProjectBabbage]

EXACTLY! Great move. Case and point, the blockchain is *not* a magical hard drive in the sky. No one is obligated to keep your things around for you, you should design an approach that doesn't require this. Make use of Overlay Services, for example.

I know you will have backlash for this, but BIG PROPS anyways, this is an important evolution for the future.

My one suggestion: bias towards keeping *unspent* outputs (or transactions + proofs that have a currently unspent state), rather than just "the most recent billion". This promotes use-cases that properly use the network's tokenization features, serving them the latest "current" tokens while eliminating the past history, in greater alignment with the white paper's approach.

A more balanced short-term approach might be "we keep all current UTXOs, plus the past 2 years of transactions" — this period can get shorter over time.

Eventually, this can transition to "We only host current UTXOs" — and you can even start to exclude certain specific, large / problematic UTXO subsets over time.

And finally, at global scale, it will be something like "we'll host UTXOs for common, public systems when it's in our interest, or you can pay us to also host your specific thing, the amount will depend on whether publicly or privately..."

That would be my two sats. Thanks for helping the BSV network evolve!

[Ian Panchèvre ᯅ @IanPanchevre]

New NBA app on Vision Pro… now you can watch live basketball as a tabletop projection.

THE FUTURE IS HERE.

[kiwi 🥝 @kiwi_251]

I am calling it now Ye will be the first celebrity that will actally sell products on the blockchain.
yeezy will be on chain

[vittorio @IterIntellectus]

everything we’ve been told about aging may be wrong

scientists just discovered that the “biological clock” tracking your aging isn’t actually controlling it, it’s just reporting the damage.

the real driver of aging are permanent dna mutations
1/

[kiwi 🥝 @kiwi_251]

@autogynefiles Being labeled to be gifted at a young age fucks up your inate reward system giving high value to the label and penalising anything that might contradict it. Its a down ward spiral of avoidance and seeking external validation until they find an equilibrum of depressing mediocrity

[JerryRigEverything @ZacksJerryRig]

I would trust DOGE more if they started at the top with the $64 billion in PPP fraud instead of instantly nuking humanitarian efforts for babies in Africa.

who does that.

[The Weeknd @theweeknd]

https://t.co/sDRRJGrshI