VC at @FirstMarkCap. Host: MAD Podcast; Organizer: Data Driven NYC, Author: MAD Landscape.
New York
Joined February 2011
- Tweets20.8K
- Following2.5K
- Followers146K
- Likes67.4K
Pinned Tweet
Crazy launch! Copycats! Massive round! On the one-month anniversary of the launch of Jev, we unpack it all with @CompleteSkeptic, CEO of @typesafeai: what Jev is, how it fixes what's broken in AI, and why the fast-followers are missing the hard part.
00:00 “95% Einstein, 5% Mr. Bean”
1:09 Why intelligence per dollar matters
3:25 Why TypeSafe built in stealth for two years
5:50 “We could have had AGI last year”
6:48 Why the product carries the marketing
9:33 What is Jev? Machine-native AI, built for software
10:53 System One vs. System Two
12:26 From simple decisions to composable AI
15:21 Automating email: giant prompts vs. smart if-statements
20:03 Probabilities, thresholds, and developer control
23:43 Why Jev is faster and cheaper than chat models
25:39 “Classification is AI-complete”
27:16 RLHF: how OpenAI trained models to please humans
31:28 Sycophancy, refusals, and the hidden costs of RLHF
34:22 RLVR: why benchmark gains don’t mean reliable automation
37:51 RLCD: optimizing for machine-native intelligence
39:54 Why TypeSafe builds its training data in-house
41:49 The Bitterest Lesson: define the task before scaling
44:39 Building Jev from open-weight models
45:56 Is Jev just a classifier? Why that’s a compliment
48:35 “Build prod, not God”
50:11 Embedding AI in software, beyond coding assistants
53:05 Jev’s copycats: matching the interface vs. the intelligence
55:33 Why benchmarks mislead the users they should help
58:37 Open weights vs. APIs: how TypeSafe weighs the tradeoff
1:00:04 Jevons paradox: a new market for machine intelligence
1:02:50 Why Diogo is skeptical of recursive self-improvement
1:04:47 Can practical automation avert an AI winter?
Matt Turck retweeted
🏎️
Today we are launching @cekuraAi's Text-to-Speech Bench. We tested 26 TTS models across 30,564 requests to compare their speed, reading accuracy, naturalness, and interruption handling.
When sent a full sentence, the fastest model started speaking in a median of 46ms, while the slowest took 1.65 seconds, nearly 36 times longer. But the fastest models were not necessarily the most natural-sounding. The model that scored highest on naturalness ranked 15th of 26 on full-sentence speed.
For the benchmark, every model read the same 104 agent sentences three times, from the same cloud region. We also tested word-by-word streaming, interruption handling, playback gaps, and request failures. Reading accuracy and naturalness were estimated using automated scorers.
We have published the complete leaderboard, methodology, and audio samples so you can compare the models yourself.
Explore the results: benchmarks.cekura.ai/text-to…
Matt Turck retweeted
We need France to return to its texhno-capitalist roots
More nuclear, more compute, more aerospace, more AI, less bureaucracy
Time for 🇫🇷/acc
Matt Turck retweeted
2/10 SurrealDB keeps gaining traction as a multi-model doc+graph+realtime database. Curious how it holds up under real distributed load, but the developer story is genuinely appealing.
github.com/surrealdb/surreal…
Matt Turck retweeted
Two new engines at 1.0. 28.3x the point reads of RocksDB in memory, 21.5M reads a second. 7.95x on disk with synchronous durability, 6.8M reads a second.
Agent Memory live at 77.2% on BEAM, 57.1% on full marks only. Every number published.
CLI 1.0, surreal-lsp, Studio 1.9 and 1.10, LangGraph.js docs, v3.3.4.
Next week: SurrealDB 4, SurrealFS, SurrealDB Earth, and more. Still shipping.
Weekly Changelog
- New in-memory storage engine, 1.0: 28.3x the point reads of RocksDB, both in memory. 21.5M reads a second
- New persistent storage engine, 1.0: 7.95x the point reads of RocksDB with synchronous durability. 6.8M reads a second. Encryption at rest on by default
- SurrealDB Agent Memory: 77.2% on BEAM at 1M tokens, 57.1% counting only full marks. Every number published
- SurrealDB CLI 1.0: one launcher for the database, Agent Memory, Cloud, extensions, and the language server
- SurrealDB Studio: support for accountless access and Agent Memory context management
- New LangGraph.js docs for Agent Memory
- SurrealDB v3.3.4
Matt Turck retweeted
Launch week wasn’t long enough.
28.3x the point reads of RocksDB, in memory.
7.95x on disk, with synchronous durability.
77.2% on BEAM at 1M tokens. 57.1% strict.
Next week: @SurrealDB 4, SurrealFS, SurrealDB Earth, and more.
Matt Turck retweeted
our CEO @CompleteSkeptic tells the tale of Dr. Einstein and Mr. Bean
> if AI responds like Einstein 95% of the time and Mr. Bean 5% of the time, how are you going to automate anything with it?
@mattturck talks us through a wild first month of Jev
Crazy launch! Copycats! Massive round! On the one-month anniversary of the launch of Jev, we unpack it all with @CompleteSkeptic, CEO of @typesafeai: what Jev is, how it fixes what's broken in AI, and why the fast-followers are missing the hard part.
00:00 “95% Einstein, 5% Mr. Bean”
1:09 Why intelligence per dollar matters
3:25 Why TypeSafe built in stealth for two years
5:50 “We could have had AGI last year”
6:48 Why the product carries the marketing
9:33 What is Jev? Machine-native AI, built for software
10:53 System One vs. System Two
12:26 From simple decisions to composable AI
15:21 Automating email: giant prompts vs. smart if-statements
20:03 Probabilities, thresholds, and developer control
23:43 Why Jev is faster and cheaper than chat models
25:39 “Classification is AI-complete”
27:16 RLHF: how OpenAI trained models to please humans
31:28 Sycophancy, refusals, and the hidden costs of RLHF
34:22 RLVR: why benchmark gains don’t mean reliable automation
37:51 RLCD: optimizing for machine-native intelligence
39:54 Why TypeSafe builds its training data in-house
41:49 The Bitterest Lesson: define the task before scaling
44:39 Building Jev from open-weight models
45:56 Is Jev just a classifier? Why that’s a compliment
48:35 “Build prod, not God”
50:11 Embedding AI in software, beyond coding assistants
53:05 Jev’s copycats: matching the interface vs. the intelligence
55:33 Why benchmarks mislead the users they should help
58:37 Open weights vs. APIs: how TypeSafe weighs the tradeoff
1:00:04 Jevons paradox: a new market for machine intelligence
1:02:50 Why Diogo is skeptical of recursive self-improvement
1:04:47 Can practical automation avert an AI winter?
Matt Turck retweeted
Agent memory benchmarks are baked.
We have just launched @SurrealDB Agent Memory, and published every number behind its score: 77.2% on BEAM at 1M, 57.1% counting only full marks.
Every strict score on this chart sits 20+ points below its headline.
surrealdb.com/blog/agent-mem…
This great conversation about Jev with @CompleteSkeptic, CEO of @typesafeai, is also available on Spotify, Apple Podcasts and here on YouTube:
youtu.be/9YsM1LPPz-A?si=lDlJ…
Matt Turck retweeted
We finally have a CRO! We're on pace to 10x in the next six months and this role was one of the missing pieces to get there.
- Joshua started his career on Wall Street
- then joined @Box and went on to become VP sales
- he was also the first GTM hire at @glean (scaled to $100M+ ARR)
We had been talking to people for months to close this role. Glad we found the best:D
Matt Turck retweeted
Fragment is building the infrastructures to make agents financial actors. Excited to be leading their Series A and partnering with @neckelthomas
fortune.com/2026/10/14/fragm…
Matt Turck retweeted
Transfer your favorite voices across languages and accents while preserving their unique timbre.
Developers can now localize existing voices to another language or regional accent while preserving its timbre, age, pace, and character.
Voice Localization is available through Gradium Studio and API.
Try it: gradium.link/voice-localizat…
Matt Turck retweeted
Our Text-to-Speech model is on the Baseten Model Library.
This means teams already running inference on Baseten can deploy Gradium TTS with one click alongside the models they already serve, with sub-100ms TTFA.
Point your agent at: docs.gradium.ai/guides/self-…
Gradium TTS by @GradiumAI is now in the Baseten Model Library, pairing ultra-low (50ms) TTFA with natural voices.
Use it with any transcription model and LLM to run your full voice agent pipeline on Baseten.
Try it here: baseten.co/library/gradium-t…
Matt Turck retweeted
We are replacing RocksDB.
Our new persistent storage engine is out today, open source under Apache 2.0.
With every commit flushed to disk: reads 7.95x faster, updates 6.10x faster, creates 26.8x faster.
Read the post: surrealdb.com/blog/we-are-re…
Speed or durability? Yes.
We're replacing RocksDB.
New persistent storage engine: reads 8.0x and updates 6.1x faster than RocksDB. Reads 73.7x faster than SlateDB (on disk, not object storage).
Run it yourself.
Stop choosing.
sdb.li/3VE7i24
Matt Turck retweeted
Espresso now optimizes Databricks Spark jobs.
Most Spark jobs run on wrong-sized machines with incorrect configs, leading to slow jobs, wasted spend, or both.
Matt Turck retweeted
In-memory or persistent? Yes.
Our new in-memory storage engine: 28.3x faster than RocksDB and 17.5x faster than LMDB for reads, both in memory. Lock-free. Embedded, in the browser or on a server.
Run it yourself.
Stop choosing.
sdb.li/4zqT3Mt
Matt Turck retweeted
@terencetao Math 2.0 Venn diagram is a positive view of how AI could expand the math frontier instead of making mathematicians depressed. Part of the need (ensure AI maths results can be explained and interpreted, in essence make Math more sustainable) have a parallel need in software. Curry-Howard :)
