investing with @blackbirdvc. always composting. love a good dad joke
Sydney, Australia
Joined October 2011
- Tweets586
- Following287
- Followers948
- Likes898
Ramp’s political attack on @airwallex is sad and pathetic.
At some point, the investors involved will realize that it’s not a flex, it’s just embarrassing.
“The campaign against Airwallex brings into stark relief how the current geopolitical climate makes it possible for a company’s adversaries to weaponize perceived ties to China against it”
theinformation.com/articles/…
Tom Humphrey retweeted
Introducing Canva Craft 🚀
Extraordinary image generation is now available to everyone, powered by our first frontier model built specifically for design.
With Craft, Canva is now the world's second most advanced AI lab for image generation on @DesignArena's global leaderboard.
See what it can do 👇
What is Australia's opportunity at the model development layer? 🇦🇺
I moderated my first podcast this week. Turns out it’s a lot like VC... ask questions that sound smart, then get out of the way and let founders do the hard work.
The Wild Hearts episode is all about the future of LLM / AI model development and the opportunity for Australia to build at this layer.
It features Pip Bingemann (Springboards.ai) and Jesse Clark (celeris.ai), representing two of the most promising early stage model development projects in Australia today.
Springboards excels at "creativity" (or entropy) while Celeris excels at "speed". And that is one of the key messages from the pod - Australia's opportunity exists at being best in the world in specialised use cases.
Key takeaways:
1. Training a model is getting easier. Pip, who is non-technical, taught himself to post-train a 4B-parameter model at midnight. Good data and a clear ideas beat brute force.
2. Specialisation is the opening. Nobody needs Australia to build the 20th best generalist LLM. Our opportunity is to be 1st in the world in something specific. Harrison.ai's model is #1 in the world at radiology, Springboards.ai's is #1 in the world for creativity, Celeris.ai's is #1 in the world for speed.
3. Opensource is crushing it. The quality of opensource models and the momentum behind them is greater than it has ever been. The future lies in post-training and model finetuning, and every company hits a scale where training its own models makes sense for cost, performance, and control. This reality is buoying the incredible growth of companies such as Baseten
4. This matters beyond the returns. This is not just about economic opportunity; this is about our sovereignty, security, and future as a country. Intelligence is like electricity, and no country wants to depend on someone else's switch. Australia has a huge opportunity staring us in the face and we need to get after it asap. We spend 95% of our time talking about building data centres; let's start building models too 🇦🇺
Hope you enjoy. Keen for feedback.
Full episode in thread:
It's pretty wild to think that Hugging Face started off as a teen chatbot, and then only took a new direction after open-sourcing the library they'd built to run it. The power of a pivot.
Agents are coming and Lorikeet is on the front foot.
Lorikeet is leaning into a new emerging channel they call "B2A" (business to agent). It enables businesses to engage the barrage of personal agents that are hitting their sites before the consumers do. Very cool.
Personal agents like Instinct and @muse are blowing up. They are flooding companies websites and support channels. This is the B2A (business to agent) era.
Today we're launching Lorikeet B2A: a suite of tools to help businesses win the B2A era.
Huge. 👏@celeris_ai
Introducing Celeris-1 Magnus. A model built for agentic work.
On τ³-bench banking, Magnus delivers 41.2% at a 55-second median, against GPT-5.6-sol's 38.1% at 79 seconds.
It’s a hybrid diffusion model derived from qwen3.8-27b, optimized for agentic workloads.
For more info, check out: celeris.ai/celeris-1-magnus
Everything I thought I knew about GTM (go to market) is changing. The new AI growth agent for marketers launched by @UseMateAI encapsulates that change. Very excited for this launch!
Meet mate, your AI growth teammate for marketers.
We spent the last 2 years in stealth building the hard part first: distribution. A live network of 12B+ shopper intent signals that tells us who’s actually in-market, right now.
Now mate puts AI on top of that network to find your customers and convert them.
It's like Claude for your marketing team but with our own consumer distribution infrastructure underneath.
We quietly signed 700 brands to the beta, drove $94.4 million in sales in 12 months for the likes of JD Sports, Brooklinen and Everlane, and are weeks away from crossing a $15M run rate.
Today, we’re thrilled to be opening the beta.
Drop a comment if you run an online store and we'll get you set up!
Huge product release from the @UnslothAI team - launching a desktop app for running local inference! 👏👏👏
Introducing Unsloth Desktop 🦥
The first desktop app to run and train models locally.
• Open-source. Runs on Mac, Windows and Linux
• Supports MLX, diffusion image/video, audio, GGUF
• Connect Claude Code and Codex to local LLMs
• 50% more accurate, self-healing tool calls + sandboxed code exec
• Works for CPU + multiGPU setups - NVIDIA, AMD, Intel, Mac
• Train models 2× faster with 70% less VRAM
• Private web search, deep research, RAG, MCP and exports (NVFP4, GGUF)
• Use Unsloth’s OpenAI-compatible API and cloud models
• Securely deploy LLMs remotely and access anywhere
Unsloth Desktop is now available on unsloth.ai and GitHub.
GitHub: github.com/unslothai/unsloth
Blog and Guide: unsloth.ai/docs/desktop
Incredible result for the @Celeris_ai team 🇦🇺. Benchmarks confirm they are the fastest AI model globally, 2x faster than the next model
Celeris-1 is now ranked as the #1 fastest AI model according to @ArtificialAnlys.
2,038 output tokens per second, #1 of all 591 models in response time, TTFT and time per intelligence task. 2x faster tokens/second than the next fastest model. In addition, we beat Grok 4.3 on HLE.
Celeris-1 proves it’s possible to deliver useful intelligence 10x faster than today's systems, without custom silicon.
Speed and latency matter. @Celeris_ai has just dropped Celeris-1, a general purpose LLM which delivers near-GPT-5 level intelligence with 15x+ faster response. To boot, this AI Lab is homegrown 🇦🇺🦘🪃, which makes me prouder than punch. Sign up at celeris.ai
Excited to launch Celeris-1 - a general purpose language model which delivers near-GPT-5 level intelligence with 15x faster response times.
Celeris-1 delivers a p50 response latency of 157ms - around 15× faster than GPT-5-mini and 17× faster than GPT-5 - while scoring comparably 76% on MMLU-Pro, compared with 78% and 81%, respectively.
The model is available starting today. Sign up at celeris.ai
The @spice_ai team been cooking! Spice is an open-source SQL engine built for data-intensive apps & AI agents - it grounds AI in enterprise data with zero ETL. Such an exciting part of the market right now
AI agents and real-time apps increasingly need analytical answers from data that was written seconds ago. Pointing those queries straight at an operational database slows production and widens security risk. The usual alternative, an ETL pipeline into a warehouse, is costly to operate and can lead to data that's stale by hours or days.
Spice 2.0 offers a new approach, and is now available.
It stands up a sandboxed analytic replica beside your operational store and keeps it current with high-throughput change data capture directly from those sources. You point Spice at a database and start running analytical queries in minutes, while production never carries the analytical load.
High-throughput CDC replication is complemented by the rest of the 2.0 platform: multi-node distributed compute on @ApacheDataFusio Ballista for petabyte-scale queries, Spice Cayenne columnar acceleration on the open @vortexdotdev format, a cluster-sidecar architecture that pairs isolated local sidecars with a shared cluster, and enterprise security enforced in the engine with OIDC, Cedar policies, PII masking, and mTLS.
"Spice is the data plane behind Summation. Every connector we ship, from Snowflake to a generic REST-as-a-table, collapses into one SQL surface, which is what makes our AI agents portable across a customer's stack. 2.0's distributed query takes the scaling concern off the table."
-Ramachandra Ramarathinam, CTO at @summation
It's open source, portable, scalable, and fast.
Read the launch blog for more details: hubs.ly/Q04p5pFw0
And, check out the thread ⬇️ for breakdowns of the major features in 2.0.
The Fable/Mythos drama makes it so blatantly clear why we need sovereign AI capabilities in Australia. Access to AI models is now defined by what passport you hold. How are we not making bigger, concerted steps toward building national capabilities?
theguardian.com/technology/2…
