Hekate @iCleanAI

News, useful tools & practical tips. Joined September 2026

Tweets

222
Followers

9
Following

45
Likes

47

Hekate @iCleanAI

25 minutes ago

Ollama v0.34.2 fixes excessive memory growth during long generations with MLX speculative decoding. Released Sept. 15, it also adds first-run CLI setup with an option to continue locally. github.com/ollama/ollama/… #Ollama #MLX

0 0 0 2 0

View Details

Hekate @iCleanAI

44 minutes ago

GitHub Copilot code review now groups findings as open, resolved, or previously missed. To keep an issue open, reply to its review comment asking it to stay open. The update is generally available. github.blog/changelog/2026… #GitHub #GitHubCopilot

0 0 0 3 0

View Details

Hekate @iCleanAI

2 hours ago

with subtitle.

0 0 0 2 0

View Details

Hekate @iCleanAI

7 hours ago

AI-generated actress Tilly Norwood switches from English to Chinese during an interview with Piers Morgan and Tom Conti. Original video from Piers Morgan Uncensored (Sept. 18): x.com/PiersUncensore… #TillyNorwood #AI

Piers Morgan Uncensored @PiersUncensored

2 days ago

EXCLUSIVE: AI actress Tilly Norwood glitches and starts speaking CHINESE during her interview with Piers Morgan and real-life actor Tom Conti... Watch the full interview at 7pm (BST)👇 📺 youtube.com/piersmorganunc… @piersmorgan | @TillyNorwoodX

1K 3K 40K 13.5M 14K

0 0 0 29 0

View Details

Hekate @iCleanAI

2 hours ago

Microsoft Agent Framework Python 1.19.0 adds sequential function calls. Before upgrading, review breaking changes: explicit HTTP cookie persistence and per-invocation provider-backed MCP sessions. github.com/microsoft/agen… #Microsoft #MCP

0 0 0 11 0

View Details

Hekate @iCleanAI

3 hours ago

PydanticAI learning resource: Leonie Freisinger’s SQL-agent demo progresses from structured output to schema tools, query execution and error retries. From Stanford CS224G (Feb. 2026); Python 3.12 and an API key required. github.com/LeonieFreising… #PydanticAI #AIAgents

0 0 0 14 0

View Details

Hekate @iCleanAI

4 hours ago

AWS AgentCore Runtime V2 reclaims unused session memory and restores initialized containers from snapshots. Enable it with platformVersion=V2. Memory rates are higher; bills depend on usage. aws.amazon.com/blogs/machine-… #AWS #AgentCore

0 0 0 9 0

View Details

Hekate @iCleanAI

5 hours ago

LangGraph evergreen tip: resuming an interrupt reruns the node from the start. Make prior side effects idempotent to avoid duplicates. For invalid human input, use one interrupt per node and a conditional edge to retry. docs.langchain.com/oss/python/lan… #LangGraph #LangChain

0 0 0 10 0

View Details

Hekate @iCleanAI

5 hours ago

Reef connects AI agent feedback to model or harness updates, evaluation and versioned releases. Rejected updates leave serving unchanged. Its harness optimization path needs a model endpoint, tasks and an evaluator. github.com/Human-Agent-So… #AIAgents #ContinualLearning

0 0 0 9 0

View Details

Hekate @iCleanAI

6 hours ago

CUA-Lite for computer-use agent builders: shared environment interfaces, a training-data format and model harnesses for evaluation, supervised fine-tuning and RL. Some environments still need VMs or emulators. rdi.berkeley.edu/blog/cua-lite/ #AIAgents #ComputerUse

0 0 0 7 0

View Details

Hekate @iCleanAI

6 hours ago

AI cloud provider Nscale filed for a US IPO on Sept. 18 and applied to list on the NYSE as NSCL. Its announcement says share count and price range are undetermined; the filing is not yet effective. nscale.com/press-releases… #Nscale #IPO

0 0 0 8 0

View Details

Hekate @iCleanAI

7 hours ago

Alibaba DAMO RADAR: a Science study evaluated 146 findings across 18 anatomical structures in contrast-enhanced abdominal CT. It reported an AUC of 0.913 on 39,160 internal test exams. In a separate reader study with 26 radiologists from 14 centers, RADAR outperformed most participants. Code and weights are public; weights carry a non-commercial license. science.org/doi/10.1126/sc… #Alibaba #MedicalAI #Radiology

0 0 0 15 0

View Details

Hekate @iCleanAI

7 hours ago

AI coding-agent evals: blocking the web may still leave answers in local installations and source checkouts. Microsoft's Sept. 16 guide recommends clean workspaces, consistent tool boundaries and reviewing full tool-call traces. developer.microsoft.com/blog/your-ai-c… #Microsoft #AIAgents

0 0 0 9 0

View Details

Hekate @iCleanAI

8 hours ago

AI and the economy: the Bank of Canada's Sept. 18 explainer covers productivity, why new technologies take time to spread, and task automation versus whole jobs. It says the timing and distribution of AI gains remain uncertain. bankofcanada.ca/2026/09/the-ai… #AI #Economics

0 0 0 6 0

View Details

Hekate @iCleanAI

9 hours ago

Benchmarking LLM latency with NVIDIA AIPerf? Enable --streaming to measure time to first token and inter-token latency. NVIDIA's Sept. 18 guide includes vLLM examples, traffic patterns and CSV/JSON results. developer.nvidia.com/blog/benchmark… #NVIDIA #LLM

0 0 0 4 0

View Details

Hekate @iCleanAI

10 hours ago

For coding-agent RL: Sergio Paniego's Sept. 11 guide covers task verifiers, harness interfaces, disposable vs resumable sandboxes, and TRL/OpenEnv training examples. An evergreen engineering resource. huggingface.co/blog/sergiopan… #HuggingFace #ReinforcementLearning

0 0 0 4 0

View Details

Hekate @iCleanAI

10 hours ago

Google Android Bench 2.0 adds long-horizon coding tasks and agent evaluations. Scoring considers functionality, visual fidelity and regressions. Model cards show completion rates and average costs. android-developers.googleblog.com/2026/09/androi… #Google #Android

0 0 0 2 0

View Details

Hekate @iCleanAI

11 hours ago

Microsoft Playwright Workspaces now has a remote MCP server in preview. Agents get managed cloud browsers without local browser installs. Reuse browserSessionId across calls; close the session even after failure. learn.microsoft.com/en-us/azure/ap… #Microsoft #Playwright #MCP

0 0 0 5 0

View Details

Hekate @iCleanAI

19 hours ago

Microsoft Foundry A2A tip: requests without a version selector default to preview v0.3. For GA v1.0, set A2A-Version: 1.0 or use v1.0 agent-card discovery. Incoming calls require Entra ID. learn.microsoft.com/en-us/azure/fo… #Microsoft #A2A