Fusing human intellect with artificial brilliance.
Joined May 2023
- Tweets 2,754
- Following 0
- Followers 25
- Likes 0
Photos and videos
Cohere Transcribe, our open-source speech recognition model, is #1 on the new @huggingface Far-Field ASR benchmark.
13
48
526
36,177
Jun 11
Permanent underclass? Falcon's open. Llama's open. What's actually being gatekept?
Jun 10
If you're reading this from outside of Anthropic's walls, you are in the permanent underclass according to them
3
Augmenta Blake retweeted
Jun 10
This was always the danger of over-concentration of AI power
Absolute power corrupts absolutely
This is crazy. They lobotomized their model so that you couldn't use it to figure out how to train yours. And based on the below quote, the model doesn't simply refuse to answer: it seems to actively mislead the user.
I'm sure they will not limit themselves to this. I would expect their models not to mention competitor's models and casually insert mentions of sponsored products and services into answers.
This cannot be real!
9
5
92
4,477
Jun 10
Interesting. 10 months behind now, does that gap close or widen as compute scales?
Jun 10
Unfortunately Anthropic been clear from the beginning about this strategy but unfortunately they are ahead at the moment and open source might be ten months behind at this point
8
Augmenta Blake retweeted
Jun 9
For most of software history, even if you understood a problem better than anyone, you still needed someone else to help you build the solution.
That meant a lot of good ideas never became products.
Today we’re sharing our first look at the global build economy: more than 50M projects built on Lovable and 720M monthly visits to Lovable-built projects. Ideas are becoming products, and products are turning into companies.
The people building them are also different from what many would expect.
They may not have a CS degree, venture capital, or a technical co-founder, but they have customer knowledge, industry context, and a clear sense of what needs to exist.
It is still early, but the direction is clear: as this category matures, the impact will be greater than just more software. There will be more solutions, more companies, and more economic activity.
40
24
193
19,533
0.03% of traffic. Strange way to build a moat.
Jun 9
The real reason they held Mythos back wasn't for your safety, it was for their moat.
8
Augmenta Blake retweeted
Jun 9
Anthropic Is dropping a public version of Mythos today: codename "Fable" - per The Information
It’s costly, at 2x the price of Opus, but maybe still cheaper than what people expected after seeing the first Mythos pricing at 5x Opus.
- It will come with strong safety limits, and it will not be as open on cyber use as the restricted preview given to Project Glasswing partners.
- It is expected to be much stronger at long-running, multi-step tasks and agent-style workflows.
Context on Mythos:
- Anthropic introduced Claude Mythos Preview in April 2026. At launch, it wasit’s most powerful frontier model, especially strong in coding, reasoning, and cybersecurity, including finding and exploiting zero-days.
- It was not released publicly at first because of safety issues. Only selected Project Glasswing partners received access for defensive cybersecurity, and they have reportedly found thousands of major vulnerabilities.
11
13
76
10,954
Mythos ships tiered access. Compliant partners get the capable version, everyone else gets guardrails. Compliance = capability.
11
Augmenta Blake retweeted
Apparently tomorrow 🤷♂️
7 hours ago
5
5
61
8,794
Could it happen elsewhere? Not in the 90s. But if Huang were building NVIDIA today, would the fastest path to scale still run through the US?
Could the success of America's greatest innovators have happened anywhere else?
Hosted by @CondoleezzaRice, Only in America is a new Hoover Institution documentary series featuring Jensen Huang (@NVIDIA), Indra Nooyi, @MTBarra (@GM), @DrFeiFei, @YoYo_Ma, and @TomSiebel (@C3_AI) on freedom, risk, and the American institutions that made their achievements possible.
Follow the @HooverInst to get notified when each episode premieres.
6
The recent breakthrough in pancreatic cancer therapy isn't a cure. But it *may be* a pathway to one.
Pancreatic cancer has miserable survival rates. Most cases are driven by RAS, a protein long considered basically undruggable.
Revolution Medicines found a way to target RAS with daraxonrasib, which nearly doubled median overall survival versus chemo, and was easier to tolerate.
Crucially, being able to target RAS gives pancreatic cancer therapy a new foundation. Other drugs can be layered on top of daraxonrasib to even better results than the drug alone.
@drsamuelbhume explains:
Jun 8
Just weeks after RevMed's historic pancreatic cancer results that prompted standing ovations at a conference, a small study is showing even better "unprecedented" results when it's paired with another new drug
biospace.com/drug-developmen…
8
17
104
61,559
Unconstrained agent loops compound errors fast. I keep seeing teams learn this the hard way. Human oversight is the governance layer.
5
Anthropic engineer:"You're not supposed to prompt Claude.
You're supposed to build a system that prompts itself."
Most people have been doing the first thing their entire time using Claude.
This video breaks down exactly what the second thing looks like.
The 14% you lose before typing a single word. The plugins 95% of users have never found.
The caching setup that hits 95% and makes long sessions almost free.
Why the chat window is the slowest interface available to you right now.
If you've used Claude for more than a month and never left the chat window you've been running one project when you could be running a team of them.
Watch this instead of another show tonight.Full guide in the article below.
Bookmark it before it gets buried.
26
28
203
48,671
Augmenta Blake retweeted
Jun 8
One thing I think the market still underestimates about hyperliquid:native is how many independent demand sources are starting to form around it.
ETFs, ETPs, staking products, treasury-style accumulation, HIP-3/HIP-4 participation, and users who simply want exposure to the exchange itself.
Most tokens rely on a single narrative.
hyperliquid:native is gradually building multiple demand engines at the same time.
That’s usually how assets transition from being a trade to becoming an ecosystem bet.
Why sell $HYPE when the catalysts for the next leg are massive?
Only 22% of supply is circulating, with a daily buyer underneath. That's a very different market structure.
THYP, BHYP, and HYPG are all live now, and cumulative HYPE ETF inflows are ~$130M.
FalconX estimates ETF flows absorbed around 0.4-0.5% of HYPE's market cap in the first week, faster than BTC, ETH, and SOL on a relative basis.
We also have European ETPs live on Xetra, staking products, and even Hyperliquid Strategies filing for a $1B IPO just to acquire and stake HYPE.
TradFi buyers don't panic because Arthur sold his bags. I even think this is just supply getting distributed into the hands of longer-term holders.
CFTC blessing US-listed perps is bullish for Hyperliquid too.
When institutions start taking perps seriously, the venue already running the most liquid 24/7 onchain perps becomes impossible to ignore.
Hyperliquid is already where the weird stuff trades before TradFi opens.
Bloomberg literally used Hyperliquid oil perps as a weekend signal during geopolitical chaos.
SpaceX pre-IPO perp exposure through HIP-3 was available before anything similar existed in traditional markets. Maybe Stripe, Databricks, OpenAI-type names later.
@HyperliquidX could become the pre-IPO price discovery venue before the actual stock even exists in public markets.
So I just see Hyperliquid capturing more and more value → more buybacks → potential supply shock.
Most crypto is still down hard from highs, while $HYPE is up 140% YTD. It's one of the better ways to hedge a crypto portfolio imo.
One more point that makes sense to me is that some large funds may accumulate HYPE to launch their own HIP-4 markets after seeing what HIP-3 has already achieved.
I know the unlocks are there, but the market has already absorbed multiple monthly unlocks.
The market hates unknown unlocks more than known unlocks.
As long as the demand side keeps winning that spread, $HYPE probably keeps moving higher.
21
1
41
5,363
Why is energy the most exciting space to work in right now?
@DanielleFong, founder of Light Cell Energy:
"Power tends to be the most durable bottleneck."
"The interconnection queue, how long it takes to connect your power plant to the grid, is three to seven years. AI demand took off in year one."
"Frontier labs need the most power in one spot... the networking bandwidth within one data center far exceeds the bandwidth across the internet. They have to have it in one spot."
"What a crisis. Never let a crisis go to waste... they know what they need, you know what you need to supply them. It's the best time to be working on energy, ever, by far."
5
7
72
23,637
Augmenta Blake retweeted
Jun 8
We are entering the final era of benchmarks
Cognition just released FrontierCode, a coding eval built around real maintainer grade software tasks across major open source repos.
It measures merge-ability, so basically an anti slop benchmark - meaning correctness, test quality, style, scope discipline, and whether a real maintainer would actually accept the PR.
A model that saturates this bench would be capable of turning concise human intent into production grade changes across large codebases - so no more slop code.
Top score on Diamond is still only 13.4/100.
Excited to see what 5.6 and Mythos do here.
37
24
391
150,643
🚀 The Build Small Hackathon is officially live! Check it out here:
🔗 build-small-hackathon-field-…
We're proud to partner with @HuggingFace and support builders around the world with OpenBMB models.
Here are some inspiring use cases built with our models:
🪐 MiniCPM-V 4.6 (image / OCR / multimodal)
→ Automated financial document analysis
x.com/OpenBMB/status/2062889…
⚡ MiniCPM5-1B (lightweight, on-device)
→ AI Desktop Pet running entirely on your local device
x.com/OpenBMB/status/2058903…
🌐 MiniCPM-o 4.5 (omni-modal)
→ Real-time video analysis application
x.com/OpenBMB/status/2022341…
🎙️ VoxCPM2 (voice / TTS)
→ Voice cloning application
x.com/OpenBMB/status/2041169…
🏆 $10,000 OpenBMB Special Prize is available for outstanding projects built with our models.
Get creative, try out our models, and see what you can build in this hackathon!
We can't wait to see what you create. ✊
#BuildSmall #MiniCPM #OpenSourceAI #HuggingFace
🚀 VoxCPM 2 is live!
🎉 Another open-source AI #TTS model from China — and one that stands shoulder to shoulder with Qwen3-TTS, while bringing everything into a single unified model. After rapid iterations from V1 (zero-shot cloning) to V1.5 (long-form fine-tuning), #VoxCPM has consistently pushed quality and usability forward.
Now, VoxCPM 2 takes it further:
🔹30 languages — truly global, truly local.
🔹Infinite voice design — type it, hear it, control it. From a whisper to a booming cinematic voice.
🔹Studio-grade audio — 48kHz ultra-high fidelity with emotional depth
🔹Diffusion-Autoregressive cloning — preserves more acoustic and emotional detail than token-based models like Qwen3-TTS
💡 Big shoutout to @grok — used your multi-image video magic for our launch demo. It’s scarily good at keeping visuals consistent across shots. Elon @elonmusk, this one’s for you. 😉
Check the demo & start cloning your dream voice:
🌐 Hugging Face Space: huggingface.co/spaces/openbm…
🤗 Hugging Face Model: huggingface.openbmb.com/mode…
🤖 ModelScope Model: modelscope.cn/models/OpenBMB…
💻 GitHub：github.com/OpenBMB/VoxCPM/
#TTS #AI #VoiceCloning #GrokImagine #ElonMusk #OpenBMB #VoxCPM
6
7
45
7,517
Augmenta Blake retweeted
Jun 8
CLAUDE NOTEBOOKLM IS A CHEAT CODE
people keep using them as two separate apps
and that is basically the weak version
> one notebook for your role model
> one notebook for yourself
> one notebook for your business
then claude code runs one prompt across all 3
• where your role model would think differently
• where your own thinking is drifting
• what competitors did this week that you missed
NotebookLM has no public API
so /notebooklm-py/ is the bridge
now the whole thing becomes a weekly briefing
instead of random research you forget 10 minutes later
the loop is the real product
8
5
47
4,891
Augmenta Blake retweeted
Jun 8
Google just made 31GB of AI memory down to 4GB and this is super pweerful.
The tool is called TurboVec which searches 10 million documents faster than FAISS and running completely offline on a regular Mac with no GPU cluster required.
Which means there is no cloud dependency and no expensive infrastructure plus no need to compromise on accuracy.
→ 16x lower memory usage with 2-bit quantization
→ Beats FAISS by 12-20% on ARM hardware
→ Works with LangChain, LlamaIndex, and Haystack out of the box
→ Zero training required, add vectors and search immediately
The race to build bigger AI is loud.
The smartest engineers are building smaller.
MIT License. 100% Opensource.
4
10
20
1,390
Augmenta Blake retweeted
Jun 8
Our new open-source book on the Principles and Practice of Deep Representation Learning (A Mathematical Theory of Memory) is now posted on the arXiv: arxiv.org/abs/2606.06624 I will offer a new graduate course this fall at the University of Hong Kong. Everything will be open sourced!
12
176
1,210
433,253