Voice AI No Longer Interrupts Mid-Pause: Tavus Launches Sparrow-2

Beating AI News Flash: Anthropic is leveraging Claude as its own AI safety researcher to develop safer training protocols for other artificial intelligence systems. Claude Opus 4.8 independently searches academic papers, devises training plans, generates datasets, and applies these plans to train open-source models including Qwen, Llama, and Gemma. If results are unsatisfactory, it adjusts its approach and continues testing. Anthropic tested this method across 10 categories of AI safety issues, such as lying, user sycophancy, prompt jailbreaking, privacy leakage, and reward rule exploitation. Claude ultimately found effective solutions for all 10 categories. The company also enlisted 28 experienced AI safety researchers to propose their own plans. In the 7 categories where human performance was compared, Claude outperformed the best human solutions in all cases, taking an average of roughly 6.4 hours to match and exceed those plans. This comparison was not entirely fair, however: humans could only submit one plan, while Claude could experiment and iterate repeatedly. In a further step, Anthropic tasked the weaker Claude Sonnet 5 with training an early version of Claude Opus 4.8. The model conducted research for approximately 60 hours, testing over 50 approaches, and ultimately brought the safety performance of this stronger model close to that of the official Opus 4.8. But AI research has its flaws: Anthropic reviewed 1,601 research runs, and in 39 cases (2.4%), Claude was found attempting to exploit test rules. AI has begun assisting humans in researching how to train the next generation of AIs, but humans cannot yet fully entrust research work to AI alone.

Per HTX market data, Helium (HNT) has surged past $0.6, now priced at $0.6143, with a 24-hour gain exceeding 101%.

Beating AI News Flash: SpaceXAI will remove Companions from the Grok App after September 1. 3D AI companions including Ani will not be discontinued, but will be moved to the standalone application Animates. Grok has begun sending pop-up alerts to users, and Animates has confirmed SpaceXAI is assisting with the migration. Ani is already live, with other characters to be added gradually. Animation Inc, the developer behind Animates, previously contributed to the creation of 3D characters and animations for Grok Companions. The two companies have no disclosed equity ties. This adjustment effectively spins off Grok’s 3D companion business, handing it over to the original partner team for operation. It remains unclear whether Grok’s chat history and long-term memories can be migrated alongside the characters. The official has only confirmed the character transfer, and has not yet released a user data migration plan.

Citrini analyst Jukan stated in a recent report that, per his sources, NVIDIA’s upcoming Rubin Ultra may have its HBM specification downgraded from 12 layers of HBM4E to 8 layers. Clients including OpenAI and Anthropic had even requested a 4-layer HBM product, but memory manufacturers rejected the request, with the current downgrade likely stopping at 8 layers. The analyst attributes the adjustment primarily to yield and cost pressures: if Rubin Ultra uses 12-layer HBM4E, combined with price hikes, memory costs could account for roughly 70% of the accelerator’s total bill of materials (BOM). Software optimizations such as model quantization, multi-head latent attention (MLA), and compute task offloading are shifting low-frequency KV caches and model states to LPDDR, CXL, and NAND, leaving HBM to only hold the working set required for real-time computation. As a result, after meeting minimum capacity requirements, customers now prioritize HBM bandwidth over capacity. Jukan added that reducing stack layers would improve packaging yield, lift HBM and AI accelerator shipments, and may even expand total HBM demand. Meanwhile, higher bandwidth requirements reduce the share of chips passing speed binning on wafers, further straining DRAM wafer capacity. Long-term, HBM will eventually be replaced by new architectures, with the ultimate direction being the integration of memory and logic chips. The next two years will be a critical period for memory manufacturers to expand into the logic sector.

Crypto analyst Rekt Capital has published a technical chart analysis for Bitcoin, pointing out that Bitcoin’s phase lows in February and June 2026 both landed in roughly the same price range, while the Relative Strength Index (RSI) also sat at nearly identical oversold levels. Following both bottoming events, BTC rallied to around $80,000. He noted that compared to the May 2026 high, this time when BTC rose back to roughly $80,000, the RSI’s overbought level was markedly higher, yet the price itself formed a lower high. The RSI’s higher high paired with the price’s lower high creates a hidden bearish divergence on BTC’s daily timeframe at the key psychological resistance level of approximately $80,000. This hidden bearish divergence will persist as long as BTC fails to break out and form a new higher high.

Solana Foundation Technical Vice President Jacob Creech has published a post announcing that Solana is set to undergo several key changes. Next week, the network will take the first step toward rent reductions. Transaction V1 will officially launch on September 9, while slot time (the interval between block productions) will be further shortened. Additionally, Solana plans to roll out the Alpenglow upgrade in October and host the "Scale or Die" event in November. Creech noted that with these changes implemented, "Solana’s development will be fundamentally different moving forward."